Strength for Today and Bright Hope for Tomorrow

Volume 3 : 11 November 2003

Editor: M. S. Thirumalai, Ph.D.
Associate Editors: B. Mallikarjun, Ph.D.
Sam Mohanlal, Ph.D.
B. A. Sharada, Ph.D.

Centre for Linguistics

Hyderabad-500 007, India


Prof. Jayaseelan

With Love and Admiration



1. The Two Lives of -unnu in Malayalam: A Response to Amritavalli and Jayaseelan M. T. Hany Babu and P.  Madhavan
2. The Order of the Inflectional Morphemes in Arabic Khaled Al-Asbahi
3. Serial Verbs With The Light Verbs ja: and de in ORIYA Bibhuti Bhusan Mahapatra
4. Multiple wh-fronting and Superiority: A Nested Movement Analysis Rahul Balusu
5. Stress and Tone in Punjabi K.G.Vijayakrishnan
6. The Prosodic Phonology of Negation in Assamese Shakuntala Mahanta
7. Phonological Awareness in Adult Illiterates: Onsets, Rimes and Analogies Abhra Jana

*** *** ***




M. T. Hany Babu

P.  Madhavan



This paper focuses on the nonpast suffix -unnu in Malayalam in the light of Amritavalli and Jayaseelan’s claim that what has been identified as tense in Dravidian is actually Aspect. Apart from providing arguments against AJ’s position, we also examine the two imperfective constructions of Malayalam, one of which is formed by the verb in -unnu form and an auxiliary verb. Our enquiry into constructions involving -unnu, also sheds light on certain fundamental properties of the two ‘be’ verbs of Malayalam aaN∂ and uNT∂ (which occur as auxiliaries in the imperfective constructions), in terms of the interpretation of noun phrases.

1. Introduction

The not so uncommon malady of mixing up meaning and form in taxonomy has handed down to us the following contradictory statements:[1]

(1)        a.    Malayalam has only two tenses, past and nonpast;

b.             -unnu is the present tense morpheme; -um the future tense morpheme, and -i/u the past tense morpheme in Malayalam.[2]

The treatment of -um as a modal suffix has resolved this paradox partly.[3] Only partly, because, it still leaves us with present and past tenses, apart from the modal suffixes, which are neither present nor past. That is, if we take (1)a literally, there should not be a present tense in Malayalam and this has led to some confusion as to the status of the “present tense morpheme” -unnu. Thus we see that in Hany Babu 1996a, it is claimed rather tentatively that “-unnu is not just a tense morpheme”, but there is no further elaboration on this matter. Our present venture, however, is not to solely make explicit what was implicit in earlier work. This paper has two roots: on the one hand, it is a response to a more recent analysis of Tense in Malayalam (and Dravidian in general) by Amritavalli and Jayaseelan (henceforth AJ); on the other hand, it endeavours to look at the suffix -unnu and constructions involving it, namely the putative “present tense” and the “present imperfect” in Malayalam, in detail and bring to light certain fundamental properties related to clause structure.[4]

AJ’s claims can be summarized as follows:

(2)        a.    There is no Tense (and, hence, no Tense Phrase) in Dravidian.

         b.   What is conventionally called Tense is actually Aspect; perfective aspect has been (wrongly) analysed as past tense and imperfective/progressive aspect as present tense.

         c.   The locus of finiteness is Mood (as instantiated in MoodP(hrase)).

In this paper we are primarily concerned with data from Malayalam and we refute the claims (2)a and (2)b. Note that the primary question that we address is not whether Dravidian has Tense Phrase, but whether Tense exists as a category in Dravidian. In AJ's analysis, what was earlier called as Tense Phrase has been reanalysed as Aspect Phrase. We argue that such a renaming process misses out on certain basic properties of the languages in question. We, however, feel that the claim (2)c cannot be dismissed like the other two. In fact, the idea that finiteness is located within the Complementizer domain has gained wide acceptance ever since Rizzi’s (1997) fine-grained analysis of the Comp­lementizer Phrase (CP). In Rizzi’s system, the Finiteness phrase (FinP) is the lowermost functional projection in the Comp­lementizer system. Rizzi does not consider the existence of a MoodP within the CP layer. If we take the MoodP posited by AJ as part of the C-system, this fits in perfectly with the general understanding about finiteness. However, we take exception to AJ’s claim that there is no Tense in Dravidian. There has indeed been a lack of clarity in understanding the Tense-Aspect system of Dravidian properly. And our effort should be seen as a step taken towards unravelling some of the mysteries of this ill-understood system. The claims that we make pertain mainly to Malayalam. It would be a fruitful enterprise, we presume, to see how this would extend to other Dravidian languages.

Let us summarize our claims as follows:

(3)         a.   Malayalam has Tense (and thereby a Tense Phrase); and the tense distinction in Malayalam is between past and nonpast (i.e. there is no present or future tense).

b.      The conjunctive participle does not essentially have a “perfective” reading.

c.       ­There are two -unnus in Malayalam: an -unnu that is the imperfective aspect morpheme, and an -unnu that occurs in generic constructions.

d.      The imperfective (nonprogressive) construction involves the aspectual -unnu and the auxiliary uNT∂, and it contrasts with the progressive construction that is formed by the auxiliary aaN∂ and the infinitive form of the verb in terms of

e.       The two imperfective forms show different properties with regard to agentivity and specificity of their subject.

2. AJ’s analysis of tense

AJ’s claim is that “Dravidian clause structure does not project a Tense Phrase”. The immediate problem they address is: what constitutes finiteness in Dravidian? Morphologically, they say, it is seen that a combination of tense and agreement is responsible for finiteness in Dravidian.  Their suggestion is that it is the presence of a Mood Phrase that is responsible for finiteness in Dravidian; and agreement morphology, according to them, is lodged in the Mood Phrase as a reflex of Indicative Mood. Identifying agreement with finiteness would, according to them, create a typological divide between Malayalam on the one hand and other major Dravidian languages like Tamil, Telugu, and Kannada on the other, since Malayalam does not exhibit agreement morphology on the verb. But if agreement morphology were a reflex of Indicative Mood, as they claim it is, positing “a null morphological reflex of Indicative Mood in Malayalam” would avoid this typological split. The obvious point that is being missed here is that, if positing a null morpheme were the solution to avoiding a typological split, one could as well posit a null agreement morpheme for Malayalam. AJ do not offer any arguments for not entertaining this solution. AJ also argue that the verbless copula sentences would be a problem for any analysis that identifies finiteness as a morphological property of the verb. But various analyses of verbless copula sentences convincingly show that the constructions without a copula are actually instances of "copula drop", since the null copula is to a large extent predictable.[5] This would mean that verbless copula clauses should have a copula in their underlying representation.

2.1 The disappearance of tense and agreement in negative sentences

One interesting problem that AJ tackle is the disappearance of tense and agreement in negative sentences in Kannada (which is also found in other Dravidian languages like Telugu and Tamil). Consider, for instance, the following pair of affirmative and negative sentences.[6] ((4) = AJ’s 1b and 3b)

(4)        a.    avanu   ban-d-anu


               ‘He came.’

         b.   avanu   bar-al(u)     illa

               he         come-inf     neg

               ‘He did not come.’

Apart from the disappearance of the agreement morpheme (-anu), the negative sentence has the main verb in the infinitive form (-al), whereas the affirmative one has the past tense morpheme (-d). There are two questions that AJ set out to answer:

(5)        a.    Since the main verb is in the infinitive form in (4)b, which element carries         finiteness?

b.   How does (4)b get past tense interpretation, given that there is no overt tense morpheme?

The answer to the first question is that it is the negative verb illa that carries finiteness in (4)b. AJ support their argument by showing that negation by illa cannot occur in “genuine nonfinite clauses, i.e. non-root gerundive and infinitive complement clauses”.  AJ also show that the form bar-al is unarguably a nonfinite form, since it typically occurs in purpose adjuncts and in control complements to verbs like try, and can also carry an overt case morpheme.  The second question is answered by positing an aspectual specification for the infinitive bar-al ‘come’. That is to say, the infinitive is specified for perfect aspect, and it is this aspectual specification that gives the sentence a past tense reading.

Another type of finite clauses where there is neither tense nor agreement is the clauses that have a modal. According to AJ’s analysis, both a modal and Agr can be the head of MoodP (which is the locus of finiteness). (Remember that Agreement, for them, is a reflex of Indicative Mood.) In the negative clauses, illa occurs in a NegP that is generated immediately below the MoodP and is raised to the MoodP (i.e. it incorporates both Negation and Mood). This solves the mystery of the complementary distribution of illa with Agr and Modals. Since there is no overt tense morpheme present in the negative sentences as well as the sentences with the modals, there need not, AJ argue, be a Tense Phrase. The complement of the MoodP would be an Aspect Phrase (AspP), which takes a VP complement. In the negative clauses, there would be a NegP between the MoodP and the AspP. AJ’s structure can be represented as: (AJ 26)

Tree diagram depicting MOOD

Neat as this solution might look, it raises another problem that is more related to theory internal questions like uniformity of clause structure. That is, as we saw in (4) above, unlike negative sentences, affirmative sentences are characterized by an overt tense morpheme. An easy solution might be to say that that is precisely the difference between affirmative and negative sentences. But this would mean that affirmative and negative clauses have different structures in Kannada. The question that worries AJ is: “why would a language have such a radical difference between its affirmative and other (negative, modal) sentences?” The solution that AJ offer is to refute the existence of a Tense Phrase (or rather, Tense as a category) first in Kannada, and then they go on to extend it to Malayalam. That is, they claim that instead of tense, what is manifested on the affirmative verb is nothing other than Aspect, and the argument they have in support of this claim is the homophony between the tense and aspect morphemes. Consider the following pair (AJ’s (1b) (repeated here from above) and (27b) respectively):

(7)        a.    avanu   ban-d-anu


               ‘He came.’

         b.   avanu   ban-d-id-d-anu

               he         come-perfect-be-past-3msg

               ‘He had come.’

In (7)b id is the auxiliary verb (past tense of iru ‘be’) and the main verb is marked for perfect aspect. AJ claim that such homophony is not accidental and that the affirmative sentence (7)a should be analysed as having an aspect morpheme rather than a tense morpheme. AJ also show that -d- also occurs as the perfect aspect marker in the perfective gerund form ban-d-addu (come-perf-nom) ‘having come’ (AJ’s (14)).  However, AJ do not draw any distinction between what they call "perfect aspect" in ban-d (the putative past tense form) and bar-al (the infinitive form) which occurs in the past negative sentences as in their (1b). Since the only difference between the two lies in the presence/absence of negation (which is brought about by the presence/absence of illa), one may safely assume that the aspect specification of the two verb forms has to be the same, which in turn leaves us wondering about the radical difference between affirmative and negative sentences.

The question, then, is: if there is no difference between the aspect specification of bar-al and ban-d what rules out the latter in negative contexts?[7] Note that it was the question about the radical difference between the structure of the clause in affirmative and negative clauses (in that there was a Tense Phrase in the former, but not in the latter) that drove AJ into this solution of positing an Aspect Phrase in the affirmative sentences instead of a Tense Phrase. We have seen that this solution does not explain why Kannada has after all decided to choose two different verb forms in affirmative and negative sentences. That is to say, positing the same structure for both the types of sentences amounts to nothing more than restating the problem. Moreover, the question about the radical difference between affirmative and negative clauses exists only insofar as one firmly holds on the position that clause structure is invariably signalled by overt morphology. In his monumental work, Cinque (1999) more or less convincingly argues for the strong position that the UG allows for just one invariant functional hierarchy in the clause structure. This means that there can be neither language specific nor construction specific variations in the clause structure. If we adopt this strong position advocated by Cinque, the question that AJ raises about the radical difference between affirmative and negative clauses in Kannada becomes informulable.

Another implicit assumption that AJ make is about the relation between overt morphology and syntax. By concluding that there can’t be a Tense Phrase in negative clauses and clauses that have a modal since they do not have overt tense morphology, AJ seem to assume that clause structure has to be signalled invariably by overt morphology. We, however, differ from AJ by arguing that lack of overt morphology is not a strong empirical argument against the existence of a Tense Phrase, in which case there ceases to be any “radical difference” between the structure of the affirmative and the negative clauses. All the same, AJ have shown convincingly that it could be the aspectual specification of the nonfinite forms that imparts a tense interpretation to the negative sentences. In the next section, we shall show with data from Malayalam that the attempt to extend AJ’s “no Tense” analysis to affirmative constructions makes it all the more untenable.

2. 2. AJ's arguments about Malayalam

AJ begin by noting that unlike Kannada, Malayalam looks straightforward in that there is no striking difference between the affirmative and negative sentences. Consider the pairs of sentences below: (AJ's 2 and 4)

(8)        a.    avan  var-unnu

   he      come-present

               ‘He comes.’

         b.   avan  van-nu[8]

               he      come-past

               ‘He came.’

(9)        a.    avan  var-unn(u)        illa

   he      come-present-  neg

               ‘He does not come.’

         b.   avan  van-n(u)     illa

               he      come-past- neg

               ‘He did not come.’

AJ reject the analysis that suggests itself: namely, the main verbs var-unnu and van-nu are marked for tense and, hence, finiteness, and illa is a negative marker (very much like not in English). The principal objection that AJ raise against this “commonsensical” analysis is that illa cannot occur in nonfinite constructions, that is, gerundive and infinitival complements in Malayalam. This restriction in the distribution of illa leads AJ to conclude that in constructions like (9) above, illa should be the finite element. This conclusion, however, does not necessarily follow from the distributional restriction of illa. The use of different strategies for finite and nonfinite negation is widely attested among the languages of the world, and the fact that illa cannot negate nonfinite verbs need only be taken as a reflex of this property in Malayalam.[9] AJ’s suggestion, however, is that illa occurs in a Mood Phrase in Malayalam (just as in Kannada), which means that it should be illa that is the finite element in (9).[10] Now, if illa is the finite element, then what about the verb forms like var-unnu and vannu? Given the line of thought AJ pursued for Kannada, one would say that they are nonfinite forms. And indeed forms like var-unnu and van-nu do occur with other finite verbs, just as their counterparts in Kannada. The nonpast var-unnu occurs in imperfective constructions with the auxiliary uNT∂. The past form is more notorious, since its occurrence as the conjunctive participle that occurs in certain aspectual as well as serial verb constructions is widely discussed in literature on Dravidian. Consider the following pair, where var-unnu and van-nu occur as verb forms marked for aspect rather than tense. (AJ’s 33) 

(10)    a.    avan  var-unn(u)        uNT

   he      come-present-  be(exis.)(present)

               ‘He is coming.’

         b.   avan  van-n(u)     irun-nu

               he      come-perf- be-past

               ‘He had come.’

In (10) uNT∂ and irun-nu are the finite verbs respectively, and hence var-unnu and van-nu have to be nonfinite. AJ convincingly argue that both these forms are marked for aspect and not for tense, but then they go on to claim that even when there is no auxiliary verb present (as in (8) above), these verbs carry aspect and not tense, and, hence, there is no need at all for a projection of tense in Malayalam. Traditionally a distinction is made between the form van-nu when it occurs as the past tense form (as in (8)b above) and when it occurs in constructions like (10)b. When it occurs in constructions like (10)b, it is called the conjunctive participle. This terminological distinction is based on the fact that the former, but not the latter, is finite. The acid test for finiteness is the use of illa negation. The negative marker illa can negate finite verbs, but not nonfinite verbs.[11] An illustrative example would be the following:

(11)    a.    avan  maaŋŋa    cetti-ttinnu

               he      mango     cut-ate

               ‘He cut the mango and ate it.’

         b.   *avan   maaŋŋa    cetti-(y)illa-tinnu

                 he       mango     cut-neg-ate

               Intended: ‘He ate the mango without cutting it.’

         c.   avan  maaŋŋa    cett-aa-te-tinnu

               he      mango     cut-neg-aug-ate

   ‘He ate the mango without cutting it.’

There are two verbs cetti ‘cut’ and tinnu ‘ate’ in (11), out of which the latter, as shown by our gloss, is in past tense. cetti is homophonous with the past tense, but it has totally different syntactic properties from the past tense in that it can only negated by the negative morpheme -aa, which characteristically occurs in nonfinite constructions, and not by illa, which negates the past tense form. In the approach outlined by AJ, finiteness is a property of the abstract Mood feature, located in the MoodP, in which case, there would be no difference between the past tense form and the conjunctive participle. Since illa has to move to MoodP, the lack of MoodP in the case of the conjunctive participle would, for them, explain the unavailability of the illa negation.

We shall look at the past tense form and the conjunctive participle in the next section try to tease out the differences between them. The -unnu form will be taken up in section 3 below.

2. 3. The past tense and the conjunctive participle: The question of perfectivity

Based on the homophony of the past tense form and the conjunctive participle, AJ argue that there is no difference between them (apart from finiteness). But homophony, a fairly widespread phenomenon, doesn’t force us to conclude that whatever is homophonous is the same. Consider the homophony of the plural morpheme of nouns and the third person singular agreement morpheme on the verbs, or the -ed in the past tense and the past participle form of the regular verbs in English. Imagine a scenario in which English did not have any irregular verbs, in which case the difference between the past participle form and the past tense form would not be visible on the surface. Following AJ, we would then be forced to analyse it as follows: the perfective is formed in English with an auxiliary verb (have) and the form of the verb marked for perfect aspect (i.e. -ed); and more importantly, there is no past tense in English, what gets interpreted as past tense is the perfect aspect of the verb.[12] It doesn’t need to be pointed out that we are indeed dealing with such a language, namely Malayalam. The aim of this thought experiment was to drive home the point that homophony need not be taken too seriously. At best, it can only be corroborative evidence, not the starting point of the hypothesis. Moreover, the claim that there is no Tense in Dravidian also raises ontological questions of wider significance, which we do not intend to address here.

         AJ claim that the conjunctive participle and the past tense form are marked for perfective aspect. We would like to show that the conjunctive participle does not essentially have “perfective” meaning. Consider the following pair of sentences:

(12)    a.    suni   waNTi     ooTicc-u pooy-i

   Suni      vehicle     drive-past   go-past

   ‘Suni went away driving a vehicle.’

         b.   abu    biiDi     waliccu    rasiccu

               Abu   beedi    smoke     enjoyed

               ‘Abu enjoyed himself, smoking a beedi.’

The first verb in both the cases in (12) is the conjunctive participle, but in neither of the instances can it be said to have a "perfective" meaning. On the contrary, what it has is an interpretation that is relative to the finite verb in the sense that the action denoted by the conjunctive participle is in some sense prior to that denoted by the finite verb. “Prior,” however, need not be taken in a strictly temporal sense. For instance, in the act of going away by driving a vehicle (as in (12)a), the act of going actually begins simultaneously with the action of driving. However, there is a “causal priority” that one can think of here, in the sense that it is the driving causes the act of going away, and not vice versa. [13] And, in fact, this is precisely the intuition captured by Rajarajavarma (1895) in one of the most celebrated traditional grammars of Malayalam, Keralapanineeyam. Raja­raja­varma treats the conjunctive participle as one of the five nonfinite forms of a Malayalam verb and calls it munvinayeccam. mun- can be translated as ‘pre-‘ or ‘prior’ and vinayeccam is a verbal or adverbial participle. A vinayeccam by definition is a nonfinite form and can occur only with a finite verb.[14] That is to say, the action denoted by the munvinayeccam, i.e., the conjunctive participle is “prior” to the action denoted by the finite verb; or in other words it can only be interpreted only as relative to the finite verb, and not absolutely.[15]

         We hope to have, thus, shown that the meaning of perfectivity cannot be attributed to the conjunctive participle. While the past tense can be said to have a perfective interpretation, the conjunctive participle is, at best, a nonfinite form of the verb that has an adverbial usage. Its temporal interpretation is dependent on the finite verb. In AJ's system where both the past tense form and the conjunctive participle have the same aspectual specification, this distinction cannot be captured. Now let us turn to the other suffix in question, namely the "present tense" marker -unnu.

3.  -unnu

From a historical perspective, it is argued that the formation of a present tense as distinct from future tense is a later development among Dravidian languages.[16] Bybee and Dahl (1989) have noted that the development of the present from present progressive is widely attested among the languages of the world.[17] The case of Malayalam seems to corroborate this observation. That is, it is likely that the -unnu was merely an imperfective aspectual marker and is in the process of acquiring the meaning of present tense. Note that we say “is in the process”, since we believe that the process is not entirely over. Contemporary Malayalam, we claim, has two types of constructions involving -unnu:

(13)     a.   an -unnu that is marked for the imperfective aspect, which can be followed by uNT∂ and negated by illa.

b.       an -unnu that is used in generic sentences, which can neither be followed by uNT∂, nor be negated by illa.

         We have already seen two types of contexts in which -unnu occurs: as the finite element and as the nonfinite (aspect) element (AJ's (2a) and (33a) respectively), which we repeat below:

(14)    a.    avan  var-unnu

   he      come-present

               ‘He comes.’

b.      avan  var-unn(u)        uNT

   he      come-present-  be(exis.)(present)

               ‘He is coming.’

The translation of (14)b clearly indicates that var-unnu has an imperfective interpretation, but what about (14)a? It is actually interesting to note that a sentence like (14)a would occur only in grammars of Malayalam as an illustrative example for the present tense. As we shall see in the discussion following, the actual use of the “present tense” in Malayalam is relatively unexplored. Asher and Kumari (1997) state that the “present tense” is used in generic statements, “for events that recur at regular intervals”, and “for an event in the very close or immediate future”. Unless modified by an adverbial, (14)a can have none of these interpretations.[18] In the next section, we shall illustrate that the -unnu in generic sentences is different from the other -unnu, but before that let us consider another of AJ’s objections to analysing -unnu as a tense morpheme:

         AJ note the occurrence of -unnu in gerunds as seen below: (AJ’s 34a)

(15)           [avan    var-unn-at]-ine                      patti   nannaL    samsaari-ccu   

    he        come-nonpast-nomin.-acc     about we           talked-past

               ‘We talked about his coming.’

AJ argue that if -unnu were a tense morpheme, one has the “embarrassment of tense inside gerunds.” The solution (again) is to say that -unnu is an aspect morpheme. The presence of tense becomes an embarrassment only if one takes "no tense within nonfinite clauses" as a gospel truth. It is indeed crosslinguistically well attested that nonfinite clauses are capable of exhibiting tense distinctions. Amritavalli (2000) mentions the case of Hebrew nominal clauses occurring in the present tense, citing Shlonsky 1997. If this is the case, it need not be an embarrassment to have tense within nonfinite clauses. All the more so, if we agree to the fact that tense is not finiteness, as AJ also conclude.

3. 1.  -unnu: The generic

In (13) above, we noted that apart from the -unnu with imperfective aspectual specification, there is another -unnu that occurs in generic sentences.[19] The use of -unnu in generic sentences seems to be a recent development in the history of Malayalam. Rajarajavarma (1895) attributes this tendency to the influence of the Indo-Aryan languages. The modal suffix -um seems to have been the only form available earlier in generic sentences. Whatever be the history, we would like to point out that -unnu and -um are not fully interchangeable in contemporary Malayalam. Consider the following pairs of sentences:

(16)    a.   chennai-yil     daivaŋŋal    tiŋŋi-ppaarkk-unnu             (Jayamohan 2001)

               Chennai-loc   gods           dense-dwell-nonpast

               ‘Gods dwell densely in Chennai.’

         b.   *chennai-yil      daivaŋŋal    tiŋŋi-ppaarkk-um

                 Chennai-loc    gods           dense-dwell-modal

(17)    a.   *ii      paatratt-il    naalu litter   veLLam   koLL-unnu

                 this  vessel-loc   four   liter    water       hold-nonpast

b.    ii     paatratt-il    naalu litter   veLLam   koLL-um[20]

       this vessel-loc   four   liter    water       hold-modal

       ‘This water can hold four liters of water.’

The contrast in the above pairs shows that judgements regarding the use of either of the generic tense forms are pretty sharp. However, it is beyond the scope of this paper to go into its intricacies.

         Hany Babu 1996b notes an interesting property about such sentences; namely, such sentences cannot be negated with illa, the usual negative marker in finite contexts. So we have the following contrast:

(18)    a.    suuryan    kizhakk∂    udikk-unnu

               sun          east            rise-nonpast

               ‘The sun rises in the east.’

         b.   *suuryan     kizhakk∂    udikk-unn-illa

                 sun           east            rise-nonpast-neg

               Intended: ‘The sun does not rise in the east.’

A correlation between the occurrence of the -unnuNT∂ and illa was also noted in Hany Babu 1996b. That is to say, -unnu form can be negated by illa, only in those cases in which -unnuNT∂ form can occur. So (18) above can be contrasted with sentences in (19) below.

(19)    a.    avan  naaLe         var-unnu-(uNT∂)

               he      tomorrow   come-nonpast-(aux)

               ‘He is coming tomorrow.’

         b.   avan  naaLe         var-unn-illa

               he      tomorrow   come-nonpast-neg

               ‘He is not coming tomorrow.’

         c.   *suuryan  kizhakk∂    udikk-unn-uNT∂

                 sun        east            rise-nonpast-aux

               Intended: ‘The Sun rises in the East.’

Thus, a generic sentence with -unnu can neither be negated with illa nor does it allow the use of the auxiliary uNT∂, while the nongeneric -unnu allows both. Hany Babu (1996b) did not go beyond capturing this fact related to distribution. We shall go one step further and claim that the -unnu that occurs in generic sentences does not have aspect specification, unlike the -unnu that occurs with the auxiliary uNT∂. We would argue that the -unnu that occurs in generic contexts actually occupies a different slot from that of the -unnu with aspectual specifications. Since the generic -unnu, the modal suffixes, and the negative illa are in complementary distribution, it stands to reason to assume that they vie for the same syntactic slot. Following the line of reasoning pursued by AJ, we shall tentatively propose that the syntactic slot in question is the MoodP. In AJ’s story the negative illa originates in the NegP and moves to the MoodP. Let us also assume that an -unnu form of the verb originates in a lower projection and moves on to the MoodP to incorporate the finiteness feature. We shall remain noncommittal about the nature of the projection in which the -unnu form originates.[21]

4. The two imperfectives of Malayalam

As we mentioned earlier, the tense-aspect system of Malayalam seems to be rather ill understood. Descriptive grammars do not go beyond a cursory listing of the various forms. Thus Asher and Kumari (1997) note that there are two imperfectives in Malayalam, but fail to notice any significant difference in meaning between the two.[22] The two imperfectives in question are:

(20)     a.   A form like var-unn uNT∂ 'is coming' (as in (14)b above), which is formed by adding the suffix -unnu to the verb stem and the addition of the auxiliary uNT∂, the existential copula.

         b.   Another form var-uka (y)aaN∂ 'is coming' which is formed by addition of the infinitival suffix -uka to the verb stem and the addition of the auxiliary aaN∂, the equative copula.

As shown in (20), these two forms translate as the present progressive in English. For the sake of convenience let us refer to the form in (20)a as the -unnuNT∂ form and the one in (20)b as the -ukayaaN∂ form. We are unaware of any attempt at teasing out the subtle distinctions between these two forms. Asher and Kumari (1997) regard the -ukayaaN∂ form as "essentially progressive", but do not consider it necessary to classify these two imperfectives further. We shall show below that this form is indeed "progressive" and differs in rather significant ways from the ­-unnuNT∂ form. The primary difference is, of course, as Asher and Kumari point out, in the intrinsic “progressive” meaning of the -ukayaaN∂ form, which makes it incompatible with stative verbs. We also find other differences: the -ukayaaN∂ form cannot be modified by an adverb of manner; certain types of arguments are excluded from its subject position. The subjects that are excluded from the subject position of the -ukayaaN∂ fall into two categories: nonagentive subject of intransitive verbs (i.e. not the subject of unaccusatives, but that of unergatives) and indefinite subjects that get a non-specific interpretation. We shall show that all these apparently disparate surface phenomena can be explained by the different properties of the auxiliary verbs in question, namely the equative copula aaN∂ versus the existential copula uNT∂

4. 1. Stative verbs

A well-known property of the stative verbs is that they can't occur in the progressive aspect. Thus forms like *She is loving him and *I am knowing that are unavailable in most native varieties of English. We can see that out of the two imperfectives we identified in (20) above, the -unnuNT∂ form can occur with stative verbs, but the -ukayaaN∂ form cannot:

(21)    a.    maňju   diliipan-e          sneehikk-unn-uNT∂

               Manju  Dileepan-acc    love-nonpast-aux

               ‘Manju loves Dileepan.’

         b.   nii      paraňň-a     kaaryam  ňaan  oorrkk-unn-uNT∂

               you    said-rel.p.   matter      I        remember-nonpast-aux

               ‘I remember what you said.’

(22)    a.    *maňju    diliipan-e          sneehikk-uka-(y)aaN

                 Manju   Dileepan-acc    love-inf-aux

               ‘*Manju is loving Dileepan.’

         b.   *nii       paraňň-a     kaaryam  ňaan  oorkk-uka-yaaN∂

                 you     said-rel.p.   matter      I        remember-inf-aux

               ‘*I am remembering what you said.’

         The fact that the stative verbs cannot occur in the ­-ukayaaN∂ form proves that we have to treat it as the progressive aspect. The -unnuNT∂ form, though it is imperfective, is not "progressive". What is common between the two forms is that both of them are used to talk about incomplete actions and states as is evident from the term “imperfective” that can be used to cover both. The sharp contrast in their behaviour with respect to stative verbs shows that they have to be further classified.[23]

4. 2. Manner adverbs

An intriguing contrast can be observed in the behaviour of the manner adverbs with respect to the two kinds of imperfectives:

(23)    a.    omana  nannaayi  paaT-unn-uNT

               Omana well         sing-nonpast-aux

               ‘Omana is singing well.’

         b.   omana     (*nannaayi)    paaT-uka-yaaN∂

               Omana    (*well)           sing-inf-aux

               ‘Omana is singing (*well).’

As shown by the ungrammaticality of the adverb of manner nannaayi ‘well’ in (23)b, the -ukayaaN∂ imperfective seems to be intolerant to modification by a manner adverb. This looks indeed puzzling. We are not aware of any correlation that has been noted in any language between Aspect and modification by adverb. One of the features that make the generative enterprise an exciting one is the possibility of unearthing connections between phenomena that are disparate on the surface.  If modification by manner adverbs is unlikely to be affected by the choice of the aspect of the verb, it has to be related to some other property of the verb. And we do indeed find such a property, to which we go directly below.[24]

4. 3. The nonagentive subject and auxiliary selection

There are certain constructions in which the syntactic subject does not get an agentive reading. The contrast between the (24) and (25) below shows that such verbs can occur in the -unnuNT∂ imperfective, but not the -ukayaaN∂:

(24)    a.    ente   vaacc∂    naTakk-unn-uNT

       my     watch      walk-nonpast-aux

       ‘My watch is working.’ (Literally ‘My watch is walking.’)

b.    aa      peena   ezhut-unn-uNT∂

       that    pen       write-nonpast-aux

       ‘That pen writes.’

(25)    a.    *ente    vaacc∂    naTakk-uka-yaaN

                 my      watch      walk-inf-aux

               ‘My watch is walking.’

b.    *aa    peena   ezhut- uka-yaaN∂

                 that  pen       write-nonpast-aux

               ‘That pen is writing.’

The English translations in (25) show that the subject obligatorily gets an agentive interpretation when -ukayaaN∂ form is used, which makes the sentences odd.[25] Though the contrast between (24) and (25) above might suggest that the choice between the -unnuNT∂ imperfective and the -ukayaaN∂ imperfective is a matter of agentive interpretation of the subject, we would like to point out that a deeper property of the language is involved here.  Consider the following construction:

(26)           awiTe   paNi  (uSaaraayi)       naTakk-uka-yaaN

               there     work (energetically)   walk-inf-aux

               ‘Work is going on very well there.’

On the face of it, (26) seems to be a counterexample to the claim that we made above about the -ukayaaN∂ form: i.e. they do not tolerate modification by manner adverbs; and they do not allow nonagentive subjects. As can be seen, the verb is in the -ukayaaN∂ form and, still, the subject paNi ‘work’ is not agentive, and a manner adverb is possible. However, a closer look would show us that such constructions are not counterexamples. That is to say, though paNi ‘work’ is the overt subject in (26), it can be shown that it is the underlying theme of naTakk-uka ‘walk’. Consider, for instance, its transitive counterpart:

(27)                 avan  paNi  naTatt-uka-yaaN

               he      work walk.caus.-inf-aux

               ‘He is managing the work.’

Our argument is that the verb naTakk-uka in (26) is an unaccusative verb. In the approach to unaccusative verbs outlined by Levin and Rappaport Hovav (1995), unaccusative verbs do not have an external argument. The D-structure object becomes the subject. So, our claim is that paNi is the internal argument at an underlying level and has moved on to the subject position. (26) should be contrasted with (25) above, where we had the construction vaacc∂ naTakk-uka (watch walk). In the latter case naTakk-uka ‘walk’ is not an unaccusative verb, but an unergative verb, which means that the argument vaacc∂ ‘watch’ is its external argument. As we saw above, this construction cannot occur in the -ukayaaN∂ form. We suggest that it is this property that distinguishes (26) from (25). That is to say, the contrast is between subjects that are external arguments at an underlying level and those that are not. What the ­-ukayaaN∂ form does not tolerate is not nonagentive subjects, but subjects that are nonagentive external arguments. We can put forth our prediction as follows:

(28)     a.   A clause with a nonagentive subject can occur only with the -unnuNT∂ imperfective, unless the verb is unaccusative.

         b.   A nonagentive internal argument (i.e. subject of an unaccusative verb) can occur in both the imperfective forms, i.e. the -ukayaaN∂ and the -unnuNT∂ imperfectives.[26]

At best, (28) is a descriptive statement. It would, again, be puzzling why the choice of the aspect interacts in this manner with the semantics of the verb or the interpretation of the subject. The explanation we offer is that this property is a reflex of the interaction of the lexical semantics of verbs and the syntactic property of the auxiliary verbs involved in these constructions.

         As we already pointed out, apart from the difference in the verb stem, the imperfective constructions also differ in the use of the auxiliary verb. The -uka form takes the auxiliary aaN∂ and the -unnu form the auxiliary uNT∂, and these two are the equative and existential copulas in Malayalam, respectively. We would like to point out that the sensitivity of auxiliary selection to the semantics of the verb is not an isolated phenomenon. Italian offers an interesting parallel to Malayalam in this regard. Consider the following pair of Italian sentences from Rosen (1984:45, as cited in Levin and Rappaport Hovav 1995). The point at issue is about the choice of the auxiliary essere ‘be’ and avere ‘have’.  Essere is taken to signal unaccusativity, while avere signals unergative status.[27]

(29)     a.   Mario   ha   continuato.  (*è)

               Mario   has continued    (is)

               ‘Mario continued.’

         b.   Il    dibattilo   è    continuato.     (*ha)

               the  debate     is    continued       (has)

               ‘The debate continued.’

Here we see that though the predicate is the same (i.e. continuato ‘continued’), the choice of the auxiliary depends on the agentive/nonagentive reading of the subject. Thus in (29)a the subject Mario gets an agentive interpretation and can be taken to be an external argument and the auxiliary chosen is ha ‘has’, while il dibattilo ‘the debate’ in (29)b would not be an external argument (i.e. an unaccusative construction) and hence the auxiliary is è ‘be’.

         Another instance where auxiliary selection is sensitive to the lexical semantics of the verb is reported by Mahapatra (2002) in Odia, where the selection of the auxiliaries ja ‘go’ and di ‘give’ seems to be determined by unaccusativity. Unaccusatives, thus, would occur with ja, while their transitive counterparts would have di as the auxiliary. Consider the following pair of sentences, where ja ‘go’ occurs with the intransitive use of cir ‘tear’, while de ‘give’ occurs with its transitive use:[28]

(30)     a.   citra-Ta:        cir-i-ja:-i-th-il-a:

               picture-class  tear-conj-go-conj-be-past-agr

               ‘The picture had (got) torn.’

         b.   ra:ma    citra-Ta:-ku         cir-i-de-i-thi-il-a:

               Rama   picture-class-to    tear-conj-give-conj-be-past-agr

               ‘Rama had torn the picture.’

         Coming back to the Malayalam example, if the subject is a nonagentive external argument, the auxiliary chosen is uNT∂, whereas if the subject is not an external argument (i.e. in the case of an unaccusative verb), there is no restriction on the auxiliary. That is to say, Malayalam seems to differ from Italian in not having any restriction on the auxiliary selection in the case of unaccusative verbs.

         The import of our discussion is that, the choice between the imperfective forms -ukayaaN∂ and -unnuNT∂ seems to be governed by deeper properties in the mapping of syntax-semantics relationships. And the fact that the latter, but not the former, can be modified by adverbs of manner seems to be a corollary of this property as can be seen from the fact that both of them go hand in hand as in (26) above, where a manner adverb is possible with -ukayaaN∂ when the subject gets a nonagentive reading. We have thus shown that our hunch that these two phenomena are related at a deeper level seems to be essentially on the right track.

We would like to wind up our discussion about agentivity and unaccusativity by discussing a type of construction brought to our notice by KJ Pramod (personal communication). Consider (31) below, where var-uka occurs with a nonagentive subject and still permits the -ukayaaN∂ construction:

(31)           avan  vallaate       moośam-aayi    var-uka-yaaN

       he      very.much   bad-become     come-inf-aux

       ‘He is becoming very weak.’

Note that var-uka is not a verb of motion here, but a verb denoting a change of state. Change of state verbs are taken to be prototypical examples of unaccusatives. Treating var-uka as an unaccusative verb here makes it fit well into our analysis, as unaccusatives would be able to occur with the -ukayaaN∂ form.

4. 4. Indefinite NPs[29]

Indefinite NPs behave in an interesting way with respect to the two kinds of imperfectives. Compare:

(32)    a.    puratt∂ or∂    aaL   nilkk-unn-uNT

               outside one    man   stand-nonpast-aux

               ‘There is a man standing outside.’

b.    *puratt∂  or∂ aaL   nilkk-uka-yaaN∂

         outside  one man   stand-inf-aux

               Intended: ‘There is a man standing outside.’

(33)    a.    or∂    aaL   puratt∂ nilkk-unn-uNT

       one    man   outside stand-nonpast      -aux

‘One (of the men) is standing outside.’/

 ‘There is a man standing outside.’

b.    or∂    aaL   puratt∂ nilkk-uka-yaaN∂

       One   man   outside stand-nonpast-aux

       ‘One (of the men) is standing outside.’

The crucial difference between the pairs of sentences in (32) and (33) is in the positioning of the indefinite or∂ aaL ‘one man’ with respect to the adverbial puratt∂ ‘outside’ and the reading of the indefinite expression. In (32)a or∂ aaL ‘one man’ occurs after the adverb and gets an existential reading, while it cannot get any interpretation in (32)b, where it occurs with the -ukayaaN∂ construction.  However, (33)b shows that the indefinite can precede the adverbial and get a partitive reading in an -ukayaaN∂ construction. Enc (1991) argues “partitives are necessarily specific,” citing evidence from Turkish, where partitives in the object position have to be obligatorily marked for accusative case. (A nonspecific object in Turkish surfaces without overt case marking.) Thus we can say that the difference in the grammaticality contrast obtained between (32)b and (33)b impinges upon specificity effect. The ambiguity of (33)a shows that both the specific and nonspecific readings are available even when the indefinite precedes the adverbial in an -unnuNT∂ construction. One may, again, wonder why the choice of aspect should interact with the interpretation of the indefinite in such a way. We would like to show that this contrast actually stems not from any difference in the intrinsic aspectual property of the two different forms in question, but from the different properties of the auxiliaries themselves.[30] Compare the following pair of sentences:

(34)    a.    puratt∂ or∂    aaL   uNT

               outside one    man   be(exist.)present

               ‘There is a man outside.’

b.    *puratt∂  or∂ aaL   aaN∂

         outside  one man   be(equat.)present

               Intended: ‘There is a man outside.’

(35)    a.    or∂    aaL   puratt∂ uNT

       one    man   outsidebe(exist.)present

‘One man is outside.’ (Also ‘One (of the men) is outside.’)

b.    or∂    aaL   puratt∂ aaN∂

       One   man   outside be(equat.)present

               ‘One (of the men) is outside.’

As is evident, the same contrast is obtained in (34) and (35), even without the presence of the -uka or -unnu forms. Thus or∂ aaL ‘one man’ in (34)a gets an existential reading while its preferred reading in (35)a is as a specific. aaN∂, as shown by the contrast between (34)b and (35)b does not tolerate a nonspecific subject.

         Let us recapitulate the differences between the two imperfective forms that we have discussed till now: 

(36)     a.   -ukayaaN∂ is the progressive and hence it cannot be used with stative verbs; while -unnuNT∂, though imperfective, is not progressive.

b.      -ukayaaN∂ is excluded in contexts where the subject is a nonagentive external argument (i.e. it is excluded in intransitives, in which the subject gets a nonagent theta role), while -unnuNT∂ has no such restrictions.

c.       -ukayaaN∂ cannot be modified by adverbs of manner, except when the subject is an internal argument, as in the case of unaccusative verbs.[31]

d.      -ukayaaN∂ cannot have a nonspecific subject, while -unnuNT∂ can.

The statement (36)a is about an intrinsic property of -ukayaaN∂, while all the rest are descriptive statements that need to be explained. The true spirit of generative enterprise would call for a unificatory analysis of these three apparently unrelated phenomena. We have already pointed out that these are indeed reducible to the interaction of the difference in the properties of the auxiliary verbs in question and the lexical semantics of the predicates. 

5.  Summary

In this paper we have reviewed AJ’s proposal that there is no Tense in Dravidian. We have put forth conceptual as well as empirical arguments against such a position. Tense as a category may not exist in Kannada negative clauses, but the question whether tense exists as a syntactic category cannot be answered by looking at just morphological facts. With respect to Malayalam, AJ have claimed that the past tense form and the nonpast -unnu forms in Malayalam are marked for aspect, and not for tense. The crucial piece of evidence they presented was the homophony of the finite and the nonfinite verb forms. Thus the past tense form is homophonous with what is traditionally called the conjunctive participle and the nonpast -unnu form occurs in certain constructions that are clearly imperfective. AJ suggest that it is the imperfective aspect that gets interpreted as present tense, and perfect aspect as the past tense. We have claimed that this may, at best, point towards the evolution of past and present tenses in the language. We have presented both syntactic as well as semantic evidence against such a reductionist treatment of tense and aspect in Malayalam. With respect to the conjunctive participle we showed that it does not necessarily have a perfective meaning. In our investigation of the -unnu form, we found that there are actually two types of constructions involving -unnu in contemporary Malayalam: the generic -unnu and the aspectual -unnu, both of which seem to differ in their syntactic as well as semantic properties. We also looked at the two imperfective constructions in Malayalam and showed that they differed in terms of aspect (that is progressive versus nonprogressive). Our investigation has brought to light certain interesting ways in which the lexical semantics of the verb interacts with aspect as well as interpretation of indefinite expressions. We have tried to show that some of these properties can be reduced to the properties of the two different auxiliary verbs, namely the existential and the equative copulas that occur with the imperfective constructions. A detailed look at the properties of the copulas involved in these constructions was however, beyond the scope of our paper.


Amritavalli, R. 2000. Kannada clause structure. In Rajendra Singh ed. The yearbook of South Asian languages and linguistics. New Delhi: Sage.

Amritavalli, R and KA Jayaseelan. (This volume). Finiteness and negation in Dravidian. CIEFL occasional papers in linguistics 10, CIEFL, Hyderabad.

Asher, RE and Kumari TC. 1997. Malayalam. New York: Routledge.

Athialy, John P. 1987. The semantics of modality: A study based on Malayalam and English. Tiruvananthapuram: Dravidian Linguistics Association.

Bybee, Joan and Oesten Dahl. 1989. The creation of tense and aspect systems in the languages of the world. Studies in Language. 13:51-103.

Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press.

Comrie, Bernard. 1985. Tense. Cambridge: Cambridge University Press.

Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic perspective. New York: Oxford University Press.

Dahl, Oesten. 1995. The episodic/generic distinction in tense-Aspect systems. In Gregory N. Carlson and Francis Jeffry Pelletier eds. The generic book. Chicago: The University of Chicago Press.

Enc, Mürvet. 1991. The semantics of specificity. Linguistic Inquiry 22:1-25.

Filip, Hana. 1999. Aspect, eventuality types and nominal reference. New York: Garland Publishing.

Finch, Shannon. 2001. Copular elements in Bengali and stage-/individual-level distinction. Talk given at SALA XXI, October 8-10, University of Konstanz.

Hany Babu, MT. 1996a. The binarity of Malayalam tense system: Some preliminary remarks. PILC Journal of Dravidic Studies 6:1-10.

Hany Babu, MT. 1996b. The syntax of Malayalam sentential negation. International Journal of Dravidian Linguistics 25:1-15.

Jayamohan. 2001. NeTumpaatayooram. Thrissur: Current Books.

Jayaseelan, KA. 1984. Control in some sentential adjuncts of Malayalam. Proceedings of the Berkeley Linguistic Society 10, 623-633. Berkeley, California: Berkeley Linguistic Society.

Kratzer, Angelika. 1989. Stage-Level and individual-level predicates. In E Bach, Angelika Kratzer and Barbara Partee eds. Papers on Quantification. NSF Report, Amherst, Massachusetts. [Also published in Gregory N. Carlson and Francis Jeffry Pelletier eds. 1995. The generic book. Chicago: The University of Chicago Press.]

Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity: At the syntax-lexical semantics interface. Cambridge, Massachusetts: The MIT Press.

Mahapatra, Bibhuti Bhushan. 1999 The four copulas in Odia. Talk given at the first Asian GLOW, CIEFL, Hyderabad.

Mahapatra, Bibhuti Bhushan. 2002. The four copulas in Odia. Manuscript, Central Institute of English and Foreign Languages, Hyderabad.

Rajasekharan Nair, N. 2002. Temporal notions in Malayalam and Tamil. Talk given at the Architecture of Grammar, January 2002, CIEFL, Hyderabad.

Rajarajavarma, AR. 1895. Keralapanineeyam. Reprinted by the Sahitya Pravarthaka Cooperative Society and distributed by National Book Stall, Kottayam.

Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman ed. Elements of Grammar. Dordrecht: Kluwer Academic.

Rosen, C. 1984. The interface between semantic roles and initial grammatical relations. In DM Perlmutter and C Rosen eds. Studies in Relational Grammar 2, Chicago, Illinois: University of Chicago Press.

Shlonsky, Ur. 1997. Clause structure and word order in Hebrew and Arabic: An essay in comparative Semitic syntax. Oxford: Oxford University Press.

Subrahmanyam, PS. 1971. Dravidian verb morphology: A comparative study. Anna­malai­nagar: Annamalai University.



[1] Though most traditional grammars make a three-way distinction in the tense system of Malayalam, one very often comes across statements about a two-way distinction in the tense system. For instance, Rajarajavarma (1895) begins the discussion on verbs by illustrating the three tense forms, goes on to claim that the present is formed by an auxiliary and hence there are only two tenses, namely past and future, and then finally states that only past tense is a “pure tense form” in Malayalam. And this is precisely the claim that we want to reiterate. See Hany Babu 1996a for further references.

[2] A statement made as recently as in 1997 by Asher and Kumari. See Asher and Kumari 1997 for a discussion of the morpho-phonological processes that determine the selection of the past tense form.

[3] See, for instance, Athialy 1987 and Hany Babu 1996a.

[4] The position that Kannada has no tense is advocated in Amritavalli 2000. Amritavalli and Jayaseelan extend the proposals to Malayalam as well.

[5] Thus, Mahapatra (1999) and Finch (2001) convincingly show the interaction of null copula with distinctions like Stage/Individual-level predicates as made by Kratzer (1989).

[6] AJ also discuss the nonpast form and its negative counterpart (in which the gerundive form of the verb occurs), which we leave out from the discussion as everything that is said about the past tense carries over to the nonpast as well. The aspectual specification that AJ attribute to the gerundive form is “imperfect”.

[7] AJ admit that their account does not say why "verb forms with similar aspectual specifications are excluded from the negative clause". See AJ, footnote 10.

[8] Actually the morphology of vannu 'came' is more complex than shown here. We are being faithful to AJ in the representation. Moreover, it does not affect the arguments in any manner.

[9] However, there is a lexical illa (the negative existential verb) that is the past equivalent of uNT∂ (the affirmative existential verb) that has nonfinite forms like the relative participle. See Hany Babu 1996b for a discussion.

[10] Actually it occurs in NegP and moves to MoodP to incorporate the finiteness feature.

[11] See footnote 9 above.

[12] Actually this seems to be a very common way in which past tense is formed in natural languages. Bybee and Dahl (1989) note that the development of past from a perfect form and present from an imperfective form is a very frequent path of grammaticalization in the languages of the world (reported in Dahl 1995).

[13] The actual temporal interpretation will depend, among other things, on the Aktionsart of the verb in question. Thus maaŋŋa cetti-ttinnu (mango cut-ate) ‘cut the mango and ate it’ as in (11) above will be interpreted as the action described by the nonfinite cetti ‘cut’ as being (in the temporal sequence) prior to the action of eating.

[14] The other type of nonfinite form is the peereccam, the form of the verb that modifies a noun (literally a name, as is evident from the word peer(∂) ‘name’). In Dravidian tradition, the peereccam is called the relative participle as it occurs in relative clauses and functions as an adjective.

[15] Comrie’s  (1976 and 1985) notion of “relative tense” could be used to capture this dependency relation between the finite and nonfinite verb. According to Comrie, the nonfinite verbs like the present participle in English exhibit relative tense, in the sense that their temporal interpretation depends on that of the finite verb.

[16] Subrahmanyam 1971:239.

[17] Reported in Dahl 1995.

[18] The presence of an overt adverbial like naaLe ‘tomorrow’ would give it the reading of something that is going to happen in the “near future” as in:

            avan  naaLe      var-unnu

            he      tomorrow            come-nonpast

            ‘He is coming tomorrow.’

[19] See also Bybee and Dahl 1989 on the development of present tenses crosslinguistically.

[20] This sentence is due to Nair (2002).

[21] It is unclear to us whether this projection should be a functional projection at all. It might as well be the VP.

[22] According to Asher and Kumari (1997:296) "[w]ithin the area of imperfective, there are separate forms that are primarily concerned with habitual actions and with continuous actions, and in the latter area, there is a form that one may regard as essentially progressive." Obviously the one that is concerned with habitual action is the -unnu form and the other one the -uka form. Asher and Kumari label the two forms as imperfective1 and imperfective2.

[23] We would like to point out that the tendency to overuse the progressive aspect (with the stative verbs, for instance) in many varieties of Indian English could be attributed to the lack of precision in identifying the subtle variations in the imperfectives available in Indian languages. Thus the English translations in (22) would be perfectly acceptable in many varieties of Indian English.

[24] A manner adverb can be forced in the -ukayaaN∂ construction, if there is a subject-oriented adverb. Consider:

            avan  karutikkuuTTi      moośamaayi     paaT-uka-yaaN∂

            he      deliberately    badly          sing-inf-aux

            ‘He is deliberately singing badly.’

We are unable to account for this contrast.

[25] The sentences in (24) are akin to the middle constructions in English. A well-known property of the middle constructions is that they are infelicitous without a manner adverb. Compare:

         i.    This knife cuts easily.

ii.       *This knife cuts.

Though a manner adverb is not obligatory in the sentences in (24), the fact that they tolerate manner adverbs, whereas the -ukayaaN∂ forms in (25) don’t, suggests that there is something parallel going on here. However, we are not able to offer a satisfactory explanation for this property at this stage.

[26] Agentivity seems to be just one of the parameters that govern the choice between the two imperfectives. While it seems to be true that nonagentive external arguments cannot occur with a verb in the -ukayaaN∂ form, whether or not the subject of an unaccusative can occur with a verb in the -ukayaaN∂ form seems to be also governed by factors like the Aktionsart of the verb. It should also be noted that causatives seem to behave differently from agentive constructions in crucial ways. A detailed investigation of all these finer points is beyond the scope of our paper.

[27] Rosen, however, does not take meaning as a criterion for the syntactic expression of the arguments.

[28] Mahapatra (personal communication) points out that the facts from Hindi are similar.

[29] The principal author would like to thank Bibhuti Bhushan Mahapatra, who has played a key role in alerting his (the author’s) mind to the kind of contrasts discussed in this section.  Many issues were discussed piecemeal over a long period of time – too long to be adequately recalled and acknowledged.

[30] However, we do not want at this stage to rule out the possibility of aspect having to play a role in the interpretation of the indefinite. We would like to draw the reader’s attention to the work by Filip (1999), where “the influence of perfective and imperfective operators on the quantificational and (in)definite interpretations of noun phrases” in Czech are discussed at length. The data that we deal with differ in that both the constructions are imperfective.

[31] But see footnote 24 above.




Khaled Al-Asbahi
University of Sana'a

1. 0. Introduction

Ouhalla (1991) has arrived at a linguistic generalization that typologically, VSO languages such as Arabic[32] can be characterized as having agreement (Agr) features inside the tense (T) features, and SVO languages such as French can be characterized as having Agr outside T.  His generalization (4) is as in (1) : 

(1)     The order of the inflectional morphemes.

"In VSO languages Agr is inside T, while in SVO languages Agr is outside T." (Ouhalla (1991 : 107).

In other words, as also stated in Ouhalla (1994), this assumption claims that in VSO-type languages T precedes Agr and in SVO-type languages Agr precedes T.  As a counter argument to this hypothesis, the current paper addresses the fact that these inflectional morphemes of Arabic cannot be in a sequential order consistently, and sometimes they are fused into one.  The paper is divided into two subsections : subjection 1.1 discusses the empirical factors, while subsection 1.2, gives the conceptual analysis. The conclusion is given in subsection 1.3.  

1. 1. The Empirical Reasons

          One strong linguistic fact that made Ouhalla (1991 & 1994) believe that in VSO languages T precedes Agr and in SVO languages Agr precedes T is the position of the future (fut.) markers in these two types of languages.  The tense element sa 'fut.' of Arabic comes initially in the verbal complex, while the same element in French, i.e., er 'fut.' comes at the final position of the verbal complex.  Consider (2a & b) and (3a & b), Ouhalla's (5) and (6). (The gloss "ind." stands for "indicator".  It does not have a syntactic value):


(2)       a.     sa  -ya    -zuur  -u     -l   ?awlaad    -u        xaal     -a     hum       

                   fut. 3sm   visit   ind. the     Nom.   uncle  Acc.  they.Gen.

b.     *   ya    -sa   -zuur  -u     -l    ?awlaad    -u       xaal     -a     hum 

                    3sm  fut.   visit   ind.  the Nom.   uncle  Acc. they.Gen.

                   ' The boys will visit their uncle.'                   


(3)       a.     ils     arriv   -er    -ont   demain  

                   they  arrive  fut.   3p    tomorrow        

            b.     *  ils    arriv  -ont   -er   demain 

                         they  arrive  3p    fut.  tomorrow   

                   ' They will arrive tomorrow.'     

(Ouhalla (1994:45))         

The Arabic example (2a), but not (2b), is grammatical, because the Agr element ya ‘3sm’ is inside the future tense sa and the verb zuur-u  ‘visit’. On the other hand, the French example (3a), but not (3b), is grammatical, because the Agr element ont '3p' is outside the tense element (i.e., er 'fut.' and the verb arriv ‘arrive’).

Ouhalla further assumes that this Arabic example compels one to revise the order of the inflectional morphemes stated by Chomsky in (4) below, so as to fit the orders of these morphemes for VSO-type languages. The intended format, i.e., (5) below, (disregarding the other phrasal categories for simplicity), accommodates the VSO languages, and Chomsky's X-bar format in (4) fits the requirements of SVO languages such as French quite well :



Chomsky, 1993 : 7



Ouhalla 1994 : 46

          Our main argument is that the ordering of T and Agr may be valid for the other languages exemplified by Ouhalla, but not for Arabic. We believe these two inflectional elements are sometimes 'fused' into one inside the verbal complex, and sometimes the order is opposite to what Ouhalla assumes. Put differently, there is no one-to-one relationship between what is known to be a 'morpheme' in the inflectional morphology of Arabic and the function it is supposed to be taking.  One cannot single out which one is which in some cases. And in some other cases Agr is in a higher position than T.    

          Let us begin with Ouhalla's examples of the verbal complex. Now, we have to leave out the modal element sa 'fut', for, if this element were the only reason for saying that T precedes Agr, then this would not be a good example, because, a typical SVO language (English) uses it in the same order as Arabic does.  This certainly is against Ouhalla's assumption. For us though sa is a tense marker, it is a modal marker as well.  The evidence for this is beyond the scope of this paper. Let us then assume that the preceding of modal elements such as the future tense will are not suitable candidates to account for the ordering of the inflectional morphemes.  We will be left with the verb ya-zuur-u ‘visit’ (see example (2a). According to Ouhalla, the prefix ya is 3sm (third person  singular  masculine),  i.e.,  agreement feature.  If ya is deleted,  the verb should remain 'imperfective' in Benmamoun's (1999b) term or 'tenseless' in Ouhalla's term.[34]  But the fact is that the remaining part of the verb after the deletion of ya will not have a lexical value in the language, or at least the base form zuur-u cannot be interpreted as present tense[35] :

(6)     *  zuur-u               

But when the prefix ya is retained, the word becomes a meaningful one : (i.e., 'he visits') 

(7)     ya- zuur-u   


          ' (he) visits '

If ya were just the Agr 3sm (i.e., the pro he), one has the right to ask as to where the morpheme which specifies that the verb is in the present form is.  Obviously, since zuur-u by itself does not indicate the present tense, the only element that carries the present form  of the verb is ya, which means that ya stands for both Agr and T.  Let us apply this concept to Ouhalla’s example (2a), disregarding the modal  sa 'fut', and adding a temporal adverb for present tense :              

(8)         ya-          zuur  -u    -l     ?awlaad  -u       xaal-    a      hum        

3sm/pres. visit  ind. the    boy.p.   Nom.   uncle.Acc.  they.Gen.         

                                                kull     -a     šahr    -in

every.Acc.  month.Gen.

' The boys visit their uncle every month.'      

In this sentence, since the verb zuur-u has nothing to say about the tense, the only element which indicates it is the prefix ya, and therefore ya is the Agr 3sm and the tense 'present' simultaneously, and this fact weakens Ouhalla's theory.  Nevertheless, we believe that our counter example is not quite clinching. In an attempt to sufficiently illustrate the fact that ya by itself clubs both Agr and T, we introduce another verb of MSA ðahab-a ‘go’ with its different inflectional morphemes in relation to the third person singular and plural agreement features, in only the present and past forms of the verb, for simplicity :       

(9)        a.    ya-ðhab-u    =       go+pres.+3sm                

             b.    ta-ðhab-u     =       go+pres.+3sf        

             c.    ya-ðhab-uun =       go+pres.+3pm               

             d.    ya-ðhab-n    =       go+pres.+3pf       

             e.    ðahab-a       =       go+past+3sm       

             f.    ðahab-at      =       go+past+3sf         

             g.    ðahab-uu     =       go+past+3pm                

             h.    ðahab-n       =       go+past+3pf        

Our arguments are as follows.  Firstly it is striking to notice that in this paradigm, ya only appears when the verb is in the present form (9a,c&d). This shows that the prefix ya has got something to do with the present tense. If so, it must be a tense morpheme for present tense, as well as an Agr element for 3sm (the morpheme has changed into [ta-] for feminine marker in (9b) ).

Secondly, another striking point that the reader must have noticed is that this morpheme is prefixed in the present form in (9a), but suffixed in the past form in (9e) (the y of ya- has undergone deletion due to some phonological constraint).  If one assumes along with Ouhalla that the prefixed elements such as [ya-] and [ta-] are Agr, the order becomes Agr > T, violating Ouhalla’s generalization (1). But, on the other hand, [-a] of [ðahab] in (9e-h) is clearly a past tense marker. Then, if the suffixed elements in the past form [-a] for masculine, and [-at] for feminine in (9e) and (9f) respectively, are taken to be Agr, then Ouhalla’s assumption becomes valid. This change in the position of the Agr makes the ordering of the inflectional elements inconsistent. Once it is Agr > T and another time T > Agr. This indicates that the Agr position is 'not fixed'.      

          Thirdly, let us concentrate on (9c&d), repeated here in (10a&b) for convenience 

(10)                a.    ya-ðhab-uun          =       go+pres.+3pm      

             b.    ya-ðhab-n              =       go+pres.+3pf  

Let us assume, yet again, along with Ouhalla that ya is an Agr. The fact that the only distinction between the masculine in (10a) and the feminine in (10b) being -uun on the former and -n on the latter suggests that -uun and n also are Agr for  gender, i.e., the former for masculine and the latter for feminine.  In addition, these two morphemes possess plural number feature.  This evidently violates Ouhalla’s rule (1); for this time, Arabic (a VSO language) shows two Agr elements split by the verb that includes T (note that we have already shown that ya is a tense as well as an Agr marker).  The order will then be Agr > T > Agr. This fact has also been observed by Shlonsky (1997:101). 

          The fact that [-n] is a feminine marker is supported by (9h). However, as (9g) suggests, the suffix -uu is a masculine marker for plural. The question as to why plural is observed as -uun in (9c) (also in (10a), but as –uu (with the deletion of -n) in (9g) is beyond the scope of this paper.[36]  For us it is sufficient to observe that Agr appears before and after T, violating Ouhalla’s assumption.    

          Finally, identifying Agr and/or T is not as easy as examined above. The bare  verb ðahab for example, holds for both Agr and T.  To illustrate this point, let us use an example from YA[37]. The equivalent word for ðahab in YA, is raah  ‘went’. In this verb Agr and T are fused (incorporated) :        

(11)       raah   ‘(he) went’       

Then, it is difficult, if not impossible, to identify the Agr and the T in this verb. The bottom line is, the vowels inside the verb not only are phonemic, but also morphemic. Compare now (11) above with (12) here (The term comm. denotes ‘command’) :

(12)    roh   ‘ go.comm’.       

The differences between (11) and (12) being only those of vowel length and vowel quality, result in the changing of the tense to past in the former and command in the latter.  Thus, changing of the Agr from third person in the former to second person in the latter strongly suggests that the vowels in the bare verb differ in quality and quantity.   

          In another expression, the quality and the quantity of the vowels inside the verb root signal inflectional and derivational morphemes. In this respect, we might adopt the classical example of MSA k t b, a ‘root’ for anything semantically related to ‘writing’.  Vowels differing in quality as well as quantity are inserted to get as many as thirty inflectional and derivational lexical items.  Some of them are listed here :

                 root                the word      inflected as  glossed as

(13)       k  t  b  ®          i)        kitaab          kutub           book/books

                             ii)       ya-ktub        katab-a        (he) writes / (he)wrote

                            the word      derived as    glossed as

              k  t  b  ®          i)        kataba                   kitaab                    he wrote/a book                                 ii)       kaatib          yu-kaatib      writer/he corresponds

For further analysis on Arabic (MSA) root see Kenstowicz (1994 : 396-450). In sum, the above facts force us to say that in Arabic, tense (T) and agreement (Agr) morphemes are fused inside the root of the verb.  The question as to which morpheme comes before or after is irrelevant, at least for Arabic. Hence if the above remarks are on the right track, we are free to adopt Chomsky's X-bar structure shown in (4) where the order is AgrS > T, or Ouhalla's structure in (5), where the order is T > Agr.     

1. 2. The Conceptual Arguments    

          Ouhalla has further reinforced his theory of TP dominating AgrSP by the fact that in Arabic the negative element la which is in the higher position than the inflectional morphemes, is always accompanied by a 'tense' morpheme, which in effect allows T to be in a higher position than AgrS in the hierarchical structure. Consider the following MSA examples. (Remember for Ouhalla and Benmamoun, present tense marker is seen as a type of 'infinitival', but for us it is a real tense (see footnote 3) ) :


(14)    a.   la       -a     yu           -saafir  -u    ar-   rajul   -u 

                Neg.pres. 3sm.pres.leave.  ind.  the  man.  Nom.      

                ' The man does not leave/is not leaving.' 

          b.   la     -n      yu-         saafir   -a     ar- rajul  -u

                Neg.fut.  3sm.pres.leave.  ind.  the man.Nom. 

                ' The man will not leave.'    

          c.   la       -m    yu-         saafir  ar-  rajul  -u           

                Neg.past  3sm.pres.leave. the man.Nom.        

                ' The man didn’t leave/has not left.'

The fact that a tense element is attached to the Neg element in (14) encourages him to claim that in Arabic, T is in fact in a higher position than AgrS. Our answer to this phenomenon comes from two angles.  Firstly, in no 'spoken' dialect of Arabic, as far as we can tell, can one find a Neg element carrying a tense marker. Only MSA uses it. Therefore it could be counted as either 'accidental', or a phenomenon only found in a 'less natural' variety of Arabic (i.e., MSA), in the sense of the generative paradigm of grammar[38].  Secondly, since the tense morphemes can be singled out clearly (i.e., [-a],  [-n] and [-m] in (14), it is quite possible that the tense morpheme of this type has cliticized onto the Neg element from somewhere in the hierarchy.  If this were not the case then how can one consider the right-cliticized NEG element (i.e., ne) in the French example in (15a) and its representation in (15b) below, following Pollock's (1989) analysis, which results in the sequence of NEG > V, just like the ones observed in (14):  

(15)    a.       pierre   dit     ne  pas    manger           

                   pierre   says  ne  NEG  to  eat   

          ' Pierre says not to eat.'            

                                                                             (Pollock (1989:413))       


Then one could strongly argue that French (an SVO-language) and Arabic (a VSO-language) behave in the same way.  This would contradict Ouhalla's generalization cited in subsection 1.1.  Hence, we believe that this piece of evidence refutes the theory of T preceding Agr.

          Further, as per Benmamoun's (1999a) analysis, tense feature in Arabic is strong. This is true because, according to Chomsky (1993 & 1995) in a language which constantly shows tense feature, (Arabic is one), T is strong, for a sentence without an overt T marker becomes ungrammatical.  On the other hand, a sentence with a mismatch of Agr is grammatical in the VS Order of Arabic and one with full agreement is ungrammatical. Compare the VS order in (16) with the SV order in (17) :


(16)    a.       ðahab    -a     l-    muhandis-uun         

                   go.past.3sm  the  engineer    .p          

          b.       * ðahab    -u     l-  muhandis-uun        

                      go.past.3pm the engineer   .p

                   'The engineers went'.      

(17)    a.       * al-    muhandis-uun  ðahab    -a                 

                     the    engineer   .p     go.past.3sm              

          b.       al-   muhandis-uun  ðahab    -u    

                   the  engineer    .p    go.past.3pm

                   'The engineers went'.      

In order to have a clear picture of subject movement on the X-bar architecture let us first redraw the tree diagram proposed by Ouhalla (5), repeated here as (18) (with changing the projection AgrP into AgrSP for accuracy) :


Since tense feature is strong we believe nominative Case is given to the subject by the head T.  (for more clarification see Chomsky 1995 : 213 footnote 11), the subject is bound to move to Spec TP via Spec AgrSP.  At Spec AgrSP the number feature on the  subject being not identical to that on the AgrS, the derivation crashes.  The movement of the subject to Spec Tp will be constrained by the intervening AgrSP projection.  Therefore, the idea of T preceding AgrS causes a conceptual problem.

We are bound to search for an alternative X-bar structure, by exchanging the AgrSP and TP positions.  That is, by TP following AgrSP as in (19) :


The SVO order is obtained by the subject first moving to the Spec TP for Case checking, then to Spec AgrSP for checking its f-features. The verb moves to AgrS stepwise via T. Both movements are appropriate. The desired Svo order is obtained.  In the Vso order since there is a mismatch between the number feature on the subject and that on the verb, the movement can only go up to Spec TP for the subject, while the verb moves on to adjoin with the head AgrS by showing [+compatible][39] number feature (i.e., identical number feature with the head AgrS).  The desired order is obtained in a principled way.

1.3  Conclusion   

          In this paper we have investigated the proposal put forward by Ouhalla (1991 & 1994) which states that in Arabic T precedes Agr.  In the analysis, we found that this proposal is inappropriate.  Subsection 1.1 exemplified that the order of these inflectional morphemes is inconsistent. That is, Agr precedes T in some situations, and T precedes Agr in some others.  In addition, a single morpheme can hold for both Agr and T simultaneously.  

          Subsection 1.2 disclosed the fact that in the VS order of MSA the Agr features on the subject and that of the head AgrS need not be identical.  It also states that the subject of Arabic checks its Nom. Case by the head T.  Hence, a clause architecture in which Agr precedes T, its argued, is more appropriate for holding the range of facts exemplified by MSA and other varieties of Arabic.           


Al-Asbahi, haled M. (2001), 'Word Order and Agreement in English and Yemeni Arabic', Ph.D. thesis, Central Institute of English and Foreign Languages, Hyderabad, India.

Benmamoun, Elabbas (1999a), 'Spec-Head Agreement and Overt Case in Arabic', in Adger, David, S. Pintzuk, B. Plunkett & G. Tsoulas (eds.) Specifiers : Minimalist approaches : (110-125), Oxford University Press, New York

Benmamoun, Elabbas (1999b), 'Arabic Morphology : The central role of the imperfective', Lingua, 108 : 175-201.

Chomsky, Noam (1993) 'A Minimalist Program for Linguistic Theory', in Hale, Kenneth & S.J. Keyser (eds.), The view from Building 20 : Essays in linguistics, in honor of Sylvian Bromberger, (1-52).  The MIT Press, Massachusetts.

Chomsky, Noam (1995), The Minimalist Program, The MIT Press, Cambridge, Massachusetts.          

Hasan, Abbas (1981), an-nahw al-waafi, Vol. I, daar-ul-maarif, Cairo.

Kenstowicz, Michael (1994), Phonology in Generative Grammar, Basil Blackwell Cambridge, Massachusetts.     

Ouhalla, Jamal (1991), Functional Categories and Parameteric variation, Routledge, London.

Ouhalla, Jamal (1994), 'Verb Movement and Word Order in Arabic', in Lightfoot, David & N. Hornstein (eds.) Verb Movement, (41-72), Cambridge, University of Maryland.

Pollock, Jean-Yves (1989), 'Verb Movement, Universal Grammar, and the Structure of IP', Linguistic Inquiry, 20 : 365-424.        

Raii, Jalal (2000), The Definite Article in Arabic and English, Ph.D. thesis, University of  Delhi, New Delhi.  

Shlonsky, Ur (1997),  Clause Structure and Word Order in Hebrew and Arabic : An essay in comparative Semitic Syntax, Oxford University Press, New York.


[32]   By Arabic we mean all varieties of the Arabic language.

[33]   MSA = Modern Standard Arabic.

[34]  The verb ya-zuur-u is a present form. However Ouhalla believes that present form is tenseless. Benmamoun (1999b) along with the traditional grammarians believes it to be a neutral verb, that is an imperfective. However for us, present tense is a tensed verb for the fact that it is capable of giving Case to the subject, just like the past form of the verb : 

   (i) ya-drus               -u-   Valiyy-un     (ii)  daras         -a-    Valiyy-un

        3sm/  ind.  Ali.   Nom.          study.past.3sm. Ali.   Nom.

        ' Ali studies/is studying '                         ' Ali studied '

[35]  One might straightforwardly argue that the form zuur-u can be taken to be a lexical entity, meaning 'you.p visit command', in Yemeni Arabic (YA) and other dialects of Arabic, with the suffix [–u] glossed as plural. In this case one might then claim that Ouhalla’s view is correct taking the fact that zuur by itself contains a T element, and the order becomes T > Agr. The  bottom line is, zuur indicates a ‘command tense’ to the second person singular masculine meaning ‘ command’.  In fact in MSA this is reduced to just zur ‘’ which clearly shows that the Agr and T are fused inside these three sounds. Yet, in (6) what we meant to argue for is the fact that zuur-u without the prefix ya- cannot be interpreted as the present tense of the verb visit.  Further, the sound /-u/ of zuur-u is an indicator, which has nothing to do with AgrS and T.

[36]  Hasan, (1981) considers it as a type of "nunation", (i.e., morphologically null sound), for the fact that it can be deleted in some contexts. For example the sound /n/ is deleted  when it is preceded by a negative element like lam 'no.past'.  Compare (i) with (ii)

   (i)    ya-  ðhab-uun  'they go'. 

  (ii)   lam ya-ðhab-uu  'they did not go'           

   In (ii) the sound /n/ is deleted, yielding that this sound may be a nunation. See Raii (2000), for the characteristics of nunation. 

[37] YA = Yemeni Arabic.

[38]   MSA does not have speakers of its own in the Arab world. Rather it is a variety used for education and mass media purposes, like the BBC English, for example.  

[39]   For a detailed analysis of [ + compatible ] number feature and subject verb movement see Al-Asbahi 2001.




Bibhuti Bhusan Mahapatra

There are verb sequences in Odia (Oriya) in which along with the main lexical verb there can be one or more auxiliary verbs. The auxiliary verbs in Odia, according to their functions, can be broadly classified into two types. The first type of auxiliary verbs appear essentially to support the tense, aspect and mood features. The second type of auxiliary verbs do not support the tense, aspect and mood features but provide various other information. In the word order, the second type of auxiliaries occur closer to the main verb than the first type.

Verb sequences occurring with the second type of auxiliaries are identified as “compound verb” constructions. (Cf. Mohanty 1992.) Alternatively such constructions are called “serial verb” constructions. (Cf. Sahoo 2001.) The auxiliary verbs in such verb sequences do not retain their usual lexical sense, and are called either “vectors” or “light verbs”.

Let us consider some examples:

1.                    citra-Ta:[1]  cir-i-ja:-i-th-il-a:

                       picture-class tear-cnp-go-pft-cop-past-agr

                       ‘The picture had (got) torn’

2.                    ra:ma citra-Ta:-ku cir-i-de-i-th-il-a:

                       Rama picture-class-to tear-cnp-give-pft-cop-past-agr

                       ‘Rama had torn the picture.’

In the above examples the auxiliary th- (be) supports the aspect and tense features. Nayak (1987: 39-40) observes that auxiliaries like th- (be) appear only if the aspect marker is present[2]. Further, “aspect markers appear only if the tense morphemes are present”. His observation suggests that copular auxiliaries like th- (be) bear both the aspect and tense features. However, the auxiliaries ja-: (go) and de- (give) in the above sentences are ‘light verbs’. About the precise functions of the auxiliaries like ja: (go) and de (give), no clear picture emerges from the accounts of Mohanty (1992) and Sahoo (2001).

In what follows, we will try to link the role of the auxiliaries ja: (go) and de (give) to different verb types and consequently to different argument structures.

1. Unaccusatives and the Auxiliary ja:

It is a standard assumption in generative grammar that the unaccusative verbs simply have a deep structure object, but they do not have a deep structure subject. (Cf. Levin and Rappaport (1995), Moro (1997), among others.) Given below are some unaccusative verbs in Odia. Let us call these verbs unaccusative type (i). The reason to group unaccusatives into different types will be made clear later:

3. unaccusatives; type (i):

                       aTak (to stop)                       libh (to get extinguished)

                baDh (to increase/grow)        luch (to get hidden)                  

                       bah (to flow)                         paD (to fall)

                       bha:s (to float)                      mis (to get mixed)

                       buD (to sink)                        pahanc (to arrive)

                       cha:D (to get detached)         pha:T (to crack)

                       Dar (to be afraid)                  phiT (to get untied)        

                       gaD (to roll)                          phuT (to blossom)

                       ghur (to revolve)                   sijh (to boil)

                       haj (to get lost)                     sukh (to dry)

                       ha:r (to get defeated)             uD (to fly)...

                       khas (to slip)

In Odia all unaccusative verbs take the verb ja:- (go) as an auxiliary. As an auxiliary ja:- does not have its lexical sense of ‘go’— (so, it is called a “light verb”). Consider the forms of the unaccusative verb libh (to get extinguished) with the auxiliary ja:-:

4.  verb   cnp   go  asp    cop    tense  agr              

     a)       libh   -i    -ja:     —      —      —      -e (gets extinguished)

     b)      libh   -i    -ga[3]    —      —      -l        -a: (got extinguished)

     c)       libh   -i    -j        —      —      -ib      -a (will get extinguished)

     d)      libh   -i    -ja:     -u       -(a)ch -ø      -i (is getting extinguished)

     e)       libh   -i    -ja:     -i        -(a)ch -ø      -i (has got extinguished)

     f)       libh   -i    -ja:     -u       -th      -il       -a: (was getting extinguished)

     g)       libh   -i    -ja:     -i        -th      -il       -a: (had got extinguished)

     h)       libh   -i    -ja:     -u       -th      -ib      -a (will be getting extinguished)

     i)        libh   -i    -ja:     -i        -th      -ib      -a (will have got extinguished)

These verb forms can occur only with their logical object. They cannot take any logical subject. Consider the following sentences:

5.                          dipa-Ta: libh-i-ga-l-a:

                             candle-class extinguish-cnp-go-past-agr

                             ‘The candle got extinguished.’

6.                          *ra:dha: dipa-Ta:-ku libh-i-ga-l-a:

                             Radha candle-class-to extinguish-cnp-go-past-agr

                             ‘Radha extinguished the candle.’

When ja:- occurs as the auxiliary of unaccusatives it indicates that the logical object is affected, undergoes some change. In the absence of ja:- the unaccusative verb simply refers to the event:

7.                          sandhya: he-l-e padma phula mauL-i-ja:-e                evening he-past-conditional lotus flower fade-cnp-go-agr

                             ‘When it is evening the lotus fades.’

8.                          ?? sandhya: he-l-e padma phula mauL-e

                             evening he-past-conditional lotus flower fade-agr

                             ‘When it is evening the lotus fades.’

In (7) the verb form says what happens to the lotus; but, in (8) the verb forms says what happens in the evening. So, there are reasons to believe that ja:- makes the predicate speak about how the logical object is affected.[4] Thus, it defines the object as the theme argument.

2. Unaccusatives and the Auxiliary de

There is another auxiliary de (give). As an auxiliary de (give) is devoid of its usual lexical sense, and functions as a “light verb”. It never occurs as the auxiliary of unaccusative verbs like libh (to get extinguished) with a simple conjunctive participle (cnp) marker:

                             Verb  cnp    give   asp    cop    tense  agr

9.                    a)   *libh  -i        -di      —      —      —      -e

                       b)   *libh  -i        -de     —      ­—      -l        -a:

                       c)   *libh  -i        -de     —      ­—      -b      -a

                       d)   *libh  -i        -de     -u       -(a)ch -ø      -i

                       e)   *libh  -i        -de     -i        -(a)ch -ø      -i

                       f)    *libh  -i        -de     -u       -th      -il       -a:

                       g)   *libh  -i        -de     -i        -th      -il       -a:

                       h)   *libh  -i        -de     -u       -th      -ib      -a

                       i)    *libh  -i        -de     -i        -th      -ib      -a

The paradigm (9) shows that unaccusative verbs do not accept de (give) as their auxiliary. However, unaccusatives like libh (to get extinguished) listed in type (i) have causative forms which are derived by adding the causative (caus) morpheme -a: to them. These causativized forms behave like transitives and can take de (give) as the auxiliary:

10.          verb  caus       cnp      de      asp    cop    tense  agr

       a)    libh   -a:   -i     -di        —      —      —      -e (extinguishes)

       b)    libh   -a:   -i     -de       —      —      -l        -a: (extinguished)

       c)    libh   -a:   -i     -de       —      —      -b      -a (will extinguish)

       d)    libh   -a:   -i     -de       -u       -(a)ch -ø      -i (is extinguishing)

       e)    libh   -a:   -i     -de       -i        -(a)ch -ø      -i (has extinguished)

       f)     libh   -a:   -i     -de       -u       -th      -il       -a: (was extinguishing)

       g)    libh   -a:   -i     -de       -i        -th      -il       -a: (had extinguished)

       h)    libh   -a:   -i     -de       -u       -th      -ib      -a (will be extinguishing)

       i)     libh   -a:   -i     -de       -i        -th      -ib      -a(will have extinguished)

Along with the logical object these verb forms take the logical subject (agent). Consider (11):

11.                        ra:dha: dipa-Ta:-ku libh-a:-i-de-l-a:

                             Radha candle-class-to extinguish-caus-cnp-give-past-agr

                             ‘Radha extinguished the candle.’

3. More Facts about Odia Unaccusatives

Let us come to some more facts which are related to the causative forms of the unaccusatives. There is a set of unaccusative verbs which do not have their transitive forms derived by the causative suffix -a:. But they have their causative counterparts which are derived— (perhaps) in a less regular/productive way. These verbs also behave like transitives. Since their morphological behaviour is different from those in type (i), let us call them type (ii). Some of these unaccusative verbs are given below:

12.                  unaccusatives; type (ii):

                       dab (to get pressed)     mar (to die)

                       jaL (to burn)                sar (to get over)

                       jhaD (to fall off) th(a:) (to be)...

                       kaT (to get cut)

Let us look at the transitive counterparts of the unaccusatives of type (ii) in (12); they are given in (13):

13.                        unaccusative                             transitive

                       a)   dab (to get pressed)—>  da:b (to press)

                       b)   jaL (to get burnt)—>       ja:L (to burn)

                       c)   jhaD (to fall off)—>       jha:D (to make something fall off)

                       d)   kaT (to get cut)—>         ka:T (to cut)

                       e)   mar (to die)—>              ma:r (to kill)

                       f)    sar (to get over)—>        sa:r (to finish)

                       g)   th(a:) (to be)—>             tho (to put/keep)

Also, there are unaccusatives whose transitive counterparts are simply lexical. Let us call them type (iii). Some examples of this type are given in (14):

14.                  unaccusatives; type (iii):

                             unaccusative                                       transitive

                       a)   dis (to appear/be seen)              dekh (to see)

                       b)   pas (to get in)                           pur (to put in)

However, the verbs in types (ii) and (iii) also obey the basic principle; that is, their unaccusative forms choose the auxiliary ja: (go) but not de (give); but their derived causative (transitive) forms choose the auxiliary de (give) but not ja: (go). Consider the forms of sar (get over) from type (ii):

15.                  a)   sar-i-ga-l-a: (got over)                *sa:r-i-ga-l-a:

                       b)   *sar-i-de-l-a:                    sa:r-i-de-l-a: (finished)

Different from the types (i)-(iii) are the verb roots which can be used both as unaccusatives and transitives. Unlike the unaccusatives in group (i)-(ii) their transitive counterparts are not derived by any morphological change in the root; e.g. by adding the causative morpheme -a: to the root; nor, as in type (iii), do they have corresponding lexical transitives to block their transitive use. They can be used both as unaccusatives and transitives without any morphological change in the root. Some of these verbs are given below as type (iv):

16.                  unaccusatives; type (iv):

                       bha:ng (to break)        buN (to spill (grains))

                       bhar (to get filled)        iD (to spill (liquid))

                       cir (to get torn)            poD (to burn/smoke) ...

                       khol (to open)

However, such roots turn out to be either unaccusatives or transitives with the choice of the auxiliary. A verb from type (iv) can choose either of the auxiliaries ja:- (go) or de (give). When the verb chooses ja:- it behaves like an unaccusative. When it chooses de (give) it behaves like a transitive. Consider the behaviour of the verb cir (get torn) from type (iv):

17.                        citra-Ta: cir-i-ja:-i-th-il-a:

                             picture-class tear-cnp-go-pft-cop-past-agr

                             ‘The picture had (got) torn’

18.                        *citra-Ta: cir-i-de-i-th-il-a:

                             picture-class tear-cnp-give-pft-cop-past-agr

                             ‘The picture had (got) torn’       (the intended sense)

19.                        *ra:ma citra-Ta:-ku cir-i-ja:-i-th-il-a:

                             Rama picture-class-to tear-cnp-go-pft-cop-past-agr

                             ‘Rama had torn the picture.’

20.                        ra:ma citra-Ta:-ku cir-i-de-i-th-il-a:

                             Rama picture-class-to tear-cnp-give-pft-cop-past-agr

                             ‘Rama had torn the picture.’

This shows that in the absence of a causative derivation of the unaccusatives, it is the choice of the auxiliary which decides the unaccusative vs transitive status of the verb. On the basis of the contrasts illustrated in (17)-(20) we propose an analysis that ja:- (go) is the unaccusative marker and de (give) is the transitive marker.

4. The Special Unaccusative he-

4.1 Multiple senses of he-      

The verb he- in Odia is a prototypical unaccusative verb. It is a functional verb which does not have a precise lexical ‘sense’. The ‘sense’ of the verb corresponds to that of English verbs like be, become, happen, come into being etc. The verb, with different aspect and tense features, forms a paradigm which is similar to that of any lexical verb. Consider the he- paradigm below:

21.                        verb   asp    aux    tense  agr

                       a)   hu      —      —      —      -e       (becomes)

                       b)   he      —      —      -l        -a:      (became)

                       c)   he      —      —      -b      -a       (will become)

                       d)   he      -u       -(a)ch -ø      -i        (is becoming)

                       e)   he      -i        -(a)ch -ø      -i        (has become)

                       f)    he      -u       -th      -il       -a:      (was becoming)

                       g)   he      -i        -th      -il       -a:      (had become)

                       h)   he      -u       -th      -ib      -a       (will be becoming)

                       i)    he      -i        -th      -ib      -a       (will have become)

As we have pointed out, the verb he- is basically a functional verb. It does not have a fixed lexical ‘sense’. However, in paradigm (21), for expository purposes we have glossed it with the sense become.

                       Like the lexical unaccusatives he- takes ja:- as the auxiliary. Look at the following he- paradigm:

22.                  V    cnp    aux    asp    cop    tense  agr

          a)          he   -i        -ja:     —      —      —      -e       (becomes)

          b)          he   -i        -ga     —      —      -l        -a:      (became)

          c)          he   -i        -j        —      —      -ib      -a       (will become)

          d)          he   -i        -ja:     -u       -(a)ch -ø      -i        (is becoming)

          e)          he   -i        -ja:     -i        -(a)ch -ø      -i        (has become)

          f)     he   -i        -ja:     -u       -th      -il       -a:      (was becoming)

          g)          he   -i        -ja:     -i        -th      -il       -a:      (had become)

          h)          he   -i        -ja:     -u       -th      -ib      -a       (will be becoming)

          i)           he   -i        -ja:     -i        -th      -ib      -a       (willhave become)

Like lexical unaccusatives he- does not accept de as the auxiliary in its non-causative form:

23.                        verb   cnp    aux    asp    cop    tense  agr

                       a)   *he    -i        -di      -ø      -ø      -ø      -e

                       b)   *he    -i        -de     -ø      -ø      -l        -a:

                       c)   *he    -i        -de     -ø      -ø      -b      -a

                       d)   *he    -i        -de     -u       -(a)ch -ø      -i

                       e)   *he    -i        -de     -i        -(a)ch -ø      -i

                       f)    *he    -i        -de     -u       -th      -il       -a:

                       g)   *he    -i        -de     -i        -th      -il       -a:

                       h)   *he    -i        -de     -u       -th      -ib      -a

                       i)    *he    -i        -de     -i        -th      -ib      -a


However, unlike lexical unaccusatives, he- cannot take the causative morpheme -a: and get causativized. So, the hypothetical form in (24) is not available in the language:

24.                        *hu-a:-i

In the absence of a derived causative form he- fails to take the auxiliary de- (give) like the lexical unaccusatives.

                       It may be observed here that de does not occur as the auxiliary of unaccusatives, but it occurs as the auxiliary of their derived causative forms which behave as transitives. This implies that de always occurs with transitives— whether derived or lexical. Conversely, the occurrence of de as an auxiliary ensures that the main verb is transitive.

4.2 he-, the Event Functor

So far as he- is concerned it behaves like lexical unaccusatives; i.e. it chooses ja: but not de. However, as a functional verb, he- still differs from the lexical unaccusatives in several ways: (i) it does not take the causative morpheme and get causativized; (ii) it does not take the auxiliary de (give) and get transitivized; (iii) above all, unlike the lexical unaccusatives it is a functional verb and can function only as an auxiliary of another predicate. This implies that he- always has the unaccusative function. As unaccusatives basically refer to ‘events’— (but not to ‘actions’ like transitives)— he- as an unaccusative refers to the ‘event’. As a functional verb it refers to the ‘event’ function.

4.3 he- as an Auxiliary

In paradigm (22) we have shown ja:- as the auxiliary of he-. But, he- itself occurs as the auxiliary of other verbs. In fact he- can occur as the auxiliary of the unaccusative libh (to get extinguished). But, the unaccusative libh should have a specific derived form to take he- as the auxiliary. With the conjunctive participle -i simply attached to the root the unaccusative does not accept he- as the aux. Look at the following forms of libh:

25.                        V       cnp    aux    asp    cop    tense  agr

                       a)   *libh  -i        -hu     —      —      —      -e

                       b    *libh  -i        -he     —      —      -l        -a:

                       c)   *libh  -i        -he     —      —      -b      -a

                       d)   *libh  -i        -he     -u       -(a)ch -ø      -i

                       e)   *libh  -i        -he     -i        -(a)ch -ø      -i

                       f)    *libh  -i        -he     -u       -th      -il       -a:

                       g)   *libh  -i        -he     -i        -th      -il       -a:

                       h)   *libh  -i        -he     -u       -th      -ib      -a

                       i)    *libh  -i        -he     -i        -th      -ib      -a

However, unaccusative roots accept the causative morpheme -a: and get causativized. The derived causative forms of the unaccusatives accept he- as an auxiliary. Look at the paradigm (26) in which he- occurs as the auxiliary of the derived causative form of the unaccusative libh (to get extinguished). The forms behave like transitives.

26.                  verb        causehe-       asp    aux    tense  agr    

     a)  libh        -a:   -hu     -ø      -ø      -ø      -e (is extinguished)

     b)  libh        -a:   -he     -ø      -ø      -l        -a: (was extinguished)

     c)  libh        -a:   -he     -ø      -ø      -b      -a (will be extinguished)

     d)  libh        -a:   -he     -u       -(a)ch -ø      -i (is being extinguished)

     e)  libh        -a:   -he     -i        -(a)ch -ø      -i (has been extinguished)

     f)  libh        -a:   -he     -u       -th      -il       -a: (was being extinguished)

     g)  libh        -a:   -he     -i        -th      -il       -a: (had been extinguished)

     h)  libh        -a:   -he     -u       -th      -ib      -a (will be being extinguished)

     i)   libh        -a:   -he     -i        -th      -ib      -a (willhave been extinguished)

In paradigm (22) we have shown ja:- as the auxiliary of he-. But, in (26), he- itself occurs as the auxiliary of unaccusative verbs. This implies that, in (26), he- does not occur as an auxiliary at the same level as that of ja:-.

                       In paradigm (4) we have shown ja:- as an auxiliary of the unaccusative verb libh (to get extinguished). In this paradigm the main unaccusative verb occurs with the conjunctive participle -i-. But in paradigm (27) the auxiliary ja:- occurs as an auxiliary with the unaccusative verb marked with the causative morpheme -a::

27.                  verb         caus   go      asp    aux    tense  agr

     a)    libh     -a:   -ja:     -ø      -ø      -ø      -e (is extinguished)

     b)    libh     -a:   -ga     -ø      -ø      -l        -a: (was extinguished)

     c)    libh     -a:   -j        -ø      -ø      -ib      -a (will be extinguished)

     d)    libh     -a:   -ja:     -u       -(a)ch -ø      -i (is being extinguished)

     e)    libh     -a:   -ja:     -i        -(a)ch -ø      -i (has been extinguish)

     f)     libh     -a:   -ja:     -u       -th      -il       -a: (was being extinguished)

     g)    libh     -a:   -ja:     -i        -th      -il       -a: (had been extinguished)

     i)     libh     -a:   -ja:     -u       -th      -ib      -a (will be being extinguished)

     J)     libh     -a:   -ja:     -i        -th      -ib      -a (willhave been extinguished)


The paradigms (26) and (27) show he- and ja:- as alternative auxiliaries of the causativized unaccusatives. However, the alternation between he- and ja:- does not seem to bring in any significant meaning difference. We suggested that the causative forms in (26) with auxiliary he- behave as transitives. The same causative forms in (27) with auxiliary ja: also behave as transitives. Consider the following sentences:

28.                        dipa-Ta:-ku libh-a:-he-l-a:

                             candle-class-to extinguish-caus-he- past-agr

                             ‘The candle was got extinguished (by somebody/ something).’

29.                        dipa-Ta:-ku libh-a:-ga-l-a:

                             candle-class-to extinguish-caus-go- past-agr

                             ‘The candle was got extinguished (by somebody/ something).’

30.                        ra:dha: dwara: dipa-Ta:-ku libh-a:-he-l-a:

                             Radha by candle-class-to extinguish-caus-he- past-agr

                             ‘The candle was got extinguished by Radha’

31.                        ra:dha: dwara: dipa-Ta:-ku libh-a:-ga-l-a:

                             Radha by candle-class-to extinguish-caus-go-past-agr

                             ‘The candle was got extinguished by Radha’

The agent argument is optionally dropped in (28)-(29). But, the agent argument is expressed by the instrumental case in (30)-(31). This suggests that the causative forms used in (28)-(31) are transitives. But, they should be considered as the passive forms of the transitive derived by the causative. As against the forms in (28)-(31), consider the forms used in (32)-(34);— the forms taken from (4):

32.                        dipa-Ta: libh-i-ga-l-a:       =(5)

                             candle-class extinguish-cnp-go-past-agr

                             ‘The candle got extinguished.’

33.                        *ra:dha dipa-Ta:-ku libh-i-ga-l-a:          =(6)

                             Radha candle-class-to extinguish-cnp-go-past-agr

                             ‘Radha extinguished the candle.’

34.                        *ra:dha: dwara: dipa-Ta:-ku libh-i-ga-l-a:

                             Radha by candle-class-to extinguish-cnp-go-past-agr

                             ‘The candle was got extinguished by Radha.’ (intended sense)

The forms cannot take the agent argument— neither with the nominative case nor with the instrumental case. So, they are not passivized transitives like the forms in (28)-(31). The form in (32) is a direct unaccusative form.

                       In fact, both he- and ja:- can occur together as the auxiliaries of the unaccusatives. The derived causative form of the unaccusative libh can take both he- and ja:- as the auxiliaries:

35.       V      caus he-     cnp    go      asp    aux    tense  agr

       a)  libh     -a:   -he     -i        -ja:     —      —      —      -e (is got                                                                                                      extingished)

       b)  libh     -a:   -he     -i        -ga     —      —      -l        -a: (was got


       c)  libh     -a:   -he     -i        -j        —      —      -ib      -a (will get


       d)  libh     -a:   -he     -i        -ja:     -u       -(a)ch -ø      -i (is being

                                                                             getting extinguished)

       e)  libh     -a:   -he     -i        -ja:     -i        -(a)ch -ø      -i (has been

                                                                             got extinguished)

       f)  libh     -a:   -he     -i        -ja:     -u       -th      -il       -a: (was being

                                                                   getting extinguished)

       g)  libh     -a:   -he     -i        -ja:     -i        -th      -il       -a: (had been

                                                                             got extinguished)

       h)  libh     -a:   -he     -i        -ja:     -u       -th      -ib      -a (will be

                                                                             getting extinguished)

       i)   libh     -a:   -he     -i        -ja:     -i        -th      -ib      -a (will have

                                                                   been got extinguished)

In (35) the causativized form of libh does not have the conjunctive participle marker. But, as showed in (36) it can occur with the conjunctive participle marker.

       V         caus   cnp    he      cnp    go      asp    cop    tense  agr

36.   a)         libh    -a:      -i        -he     -i        -ja:     —      —      —      -e (gets


  b)  libh       -a:      -i        -he     -i        -ga     —      —      -l        -a: (got


  c)  libh       -a:      -i        -he     -i        -j        —      —      -ib      -a (will

                                                                   get extinguished

  d)  libh       -a:      -i        -he     -i        -ja:     -u       -(a)ch -ø      -i (is

                                                                   being extinguished)

  e)  libh       -a:      -i        -he     -i        -ja:     -i        -(a)ch -ø      -i (has

                                                                   been extinguished)

  f)  libh       -a:      -i        -he     -i        -ja:     -u       -th      -il       -a: (was

                                                                   being extinguished)

  g)  libh       -a:      -i        -he     -i        -ja:     -i        -th      -il       -a: (had

                                                                   been extinguished)

  h)  libh       -a:      -i        -he     -i        -ja:     -u       -th      -ib      -a (will

                                                                   being extinguished)

  i)   libh       -a:      -i        -he     -i        -ja:     -i        -th      -ib      -a (will

                                                          have been extinguished)

From the unaccusative paradigms discussed above one can assume that the verb ja: (go) can occur as the auxiliary with both the non-causative and the causativized forms of the unaccusatives. However, he- always occurs as the auxiliary of the causativized forms of the unaccusatives.

                       So far our description of the unaccusative verb paradigms shows that both he- and de- can alternatively occur as the auxiliaries of the derived causative forms of the unaccusatives. However, they can occur together as auxiliaries. Look at the following paradigm of libh (to get extinguished):

37.                        verb   caus   cnp    de      nom   he      asp    cop    tense  agr

                       a)   libh    -a:      -i        -di      -a:      -hu     —      —      —      -e (is extinguished)

                       b)   libh    -a:      -i        -di      -a:      -he     —      —      -l        -a: (was extinguished)

                       c)   libh    -a:      -i        -di      -a:      -he     —      —      -b      -a (will be extinguished)

                       d)   libh    -a:      -i        -di      -a:      -he     -u       -(a)ch -ø      -i (is being extinguished)

                       e)   libh    -a:      -i        -di      -a:      -he     -i        -(a)ch -ø      -i (has been extinguished)

                       f)    libh    -a:      -i        -di      -a:      -he     -u       -th      -il       -a: (was being extinguished)

                       g)   libh    -a:      -i        -di      -a:      -he     -i        -th      -il       -a: (had been extinguished)

                       h)   libh    -a:      -i        -di      -a:      -he     -u       -th      -ib      -a (will being extinguished)

                       i)    libh    -a:      -i        -di      -a:      -he     -i        -th      -ib      -a (will have been extinguished)

The paradigm shows the relative positions of de- and he-. Since de- occurs at a higher level than he- it does not occur as an auxiliary of he-.

                       The following paradigm further shows the relative positions of de-, he- and ja:-:


    verb            caus         cnp    de      nom   he      cnp    go      asp    cop  tense  agr

a) libh             -a:   -i        -di      -a:      -he     -i        -ja:     —      —      —      -e

(is extinguished)

b) libh             -a:   -i        -di      -a:      -he     -i        -ga     —      —      -l      -a:

(was extinguished)

c) libh             -a:   -i        -di      -a:      -he     -i        -j        —      —      -ib      -a

(will be extinguished)

d) libh             -a:   -i        -di      -a:      -he     -i        -ja:     -u       -(a)ch -ø      -i

(is being extinguished)

e) libh             -a:   -i        -di      -a:      -he     -i        -ja:     -i        -(a)ch -ø      -i (has been extinguished)

f) libh              -a:   -i        -di      -a:      -he     -i        -ja:     -u       -th      -il -a: (was being extinguished)

g) libh             -a:   -i        -di      -a:      -he     -i        -ja:     -i        -th      -il -a: (had been extinguished)

h) libh             -a:   -i        -di      -a:      -he     -i        -ja:     -u       -th      -ib      -a (will being extinguished)

i) libh              -a:   -i        -di      -a:      -he     -i        -ja:     -i        -th      -ib      -a (will have been extinguished)

Our analysis suggests that ja: (go) is the unaccusative marker, and he- is a functional verb which always has the unaccusative function. However, further investigation is required about the precise function of he- and how he- is functionally different from ja:- (go).


Levin, B and Rappaport, M. 1995. Unaccusativity: At the Syntax-Lexical Semantics Interface.  MIT press.

Mahapatra, B. B. 2002. Stage-level vs Individual-level Predicates and the Four Copulas in Odia. Doctoral Dissertation submitted to CIEFL, Hyderabad.

Mohanty, G. 1992. Compound Verbs in Oriya. Doctoral dissertation, University of Pune.

Moro, Andrea 1997. The Raising of Predicates. Cambridge University Press.

Nayak, Rath 1987. Non-finite Clauses in Oriya and English. Doctoral dissertation, CIEFL, Hyderabad.

Sahoo, K. 1999. ‘The Syntax of Definiteness in Oriya.’ In The Year Book of South Asian Languages and Linguistics. (ed.) R. Singh. New Delhi: Thousand Oaks/California: Sage Publications.

- - - - - - 2001. Oriya Verb Morphology and Complex Verb Constructions, Doctoral Dissertation, University of Science and Technology, Trondheim, Norway.


[1] The morpheme -Ta: is a classifier available in the DP system of Odia. It never occurs as an independent morpheme. When it occurs prenominally it is suffixed either to a numeral or a quantifier, and the NP has an indefinite reading. When it occurs postnominally the NP has a definite reading. Like other Indian languages, Odia does not have a definite marker like the English the. In the absence of a demonstrative, the postnominal occurrence of -Ta: gives the definite reading to the NP. (See Sahoo (1999) for some details.)

[2] The fact is not different from English where the auxiliaries have or be appear only if the next verb has the aspect marker -en or -ing.

[3] ja: has an allomorph ga which occurs only in the simple past.

[4] Mahapatra (2002) discusses this point with more details.





I Introduction

Research into the question  What moves where when in which language and why ?  has become very interesting with the increasing amount of evidence from languages other than English. One of the puzzles for movement theories is the multiple wh-fronting exhibited by some languages (e.g. Bulgarian and Malayalam). Consider examples (1) and (2)

(1)     kaj      kakvo     nakogo     e dal ?                          (Bulgarian)

who     what       to whom    has given

‘Who gave what to whom ?’

(2)     aar∂            aarkk∂       ent∂      koTuttu ?               (Malayalam)

          who.nom    who.dat      what     give-past

          ‘Who gave what to whom ?’

Deriving the correct order of the fronted wh-words using feature attraction as in current syntactic theory poses a problem. In languages like English which front only one wh-word it is always the highest question word which gets fronted.

(3)               Who gave what to whom ?

Mutiple Wh-

This leads to the Superiority condition (Chomsky ’73) which is a fairly robust condition and holds in many cases. But the problem with (1) is that following Superiority the derivation should be like this:

Mutiple Wh-

The series of nested movements give the word order   IO  DO  S  among the question words.  But the surface word order is   S   DO   IO   , the exact opposite of the word order expected when the movements obey Superiority.

II Anti-Superiority

          One of the solutions proposed for this problem is Watanabe’s (1992)  Anti-Superiority condition:  ‘A multiple question is well formed only if there is a wh-phrase which is not c-commanded by the wh-phrase that is moved first.’  It boils down to saying that the lowest wh-word must move first

Mutiple Wh-

The problem with this solution is that it is a stipulation which makes English and Japanese obey two opposing conditions. If the anti-superiority condition applies in English it would derive the wrong word order. It is therefore limited to languages like Japanese.

III  Tuck-in

          A more recent proposal to tackle the problem of Superiority is Norvin Richard’s (1997) idea of Tuck-in. According to this all movements obey the principle of Shortest, i.e. the moved element moves the shortest distance in terms of the number of nodes it crosses. Consider (5)

Mutiple Wh-

Now, if the DO moves above S, the distance between it and the attracting head will have S in it.

Multiple Wh-

However if the DO ‘tucks-in’ between the S and C, the distance will not cross S and will therefore be the shortest.

Multiple Wh-

The order will now be

Multiple Wh-

The same logic applies to the IO

Multiple Wh-

This derives the desired word order of languages like Bulgarian, Japanese etc.

One of the problems with this proposal is that when an element ‘tucks-in’ it necessitates the creation of another intermediate position by ‘stretching’ the space between the original maximal projection and the intermediate projection.

Multiple Wh-

This proposal also weakens the Extension condition which says that any tree building operation should occur at the top of the tree i.e. it should extend the tree further up. But Tuck-in does precisely the opposite. Successive movement operations target positions successively lower in the tree.

Multiple Wh-


Another problem with this proposal is that it uses multiple specifiers to get it’s results, which are possible only in the current Chomskyean framework, not in the system using Kayne’s (1994) antisymmetry which does not allow multiple specifiers because they cannot be linearized. So this proposal does not work in a Kaynean system.

          The problem, in a nutshell, is how to get a crossed set of movements that replicate the base order and obey the Superiority condition.

IV.   Double Nested Movements

My solution to the problem is this: what seems like a crossed set of movements is actually 2 sets of nested movements ( which obey Superiority) resulting in the mirroring of the base order.

Multiple Wh-


In a system following Chomsky’s (2001) Derivation by Phase proposal, all the material that needs to be accessible outside the vP moves to the left edge of the vP. The wh-phrases which are arguments are base generated within the vP, then they move to the left edge of the vP as in (7):

Multiple Wh-

When the derivation proceeds to the next higher phase i.e. the CP there is a C with a [+P] or [+wh] feature which attracts the wh-phrases to it’s specifier positions from the specifier positions of the vP as in (8):

Multiple Wh-

The final order of the wh-phrases in the COMP is S  DO  IO,  which is what is observed in multiple wh fronting languages like Bulgarian.  So by using two already existing series of movements we derive the desired surface order without any further principles or conditions on movement.

          This is alright in a Chomskyean system using multiple specifier positions but how do we derive this in a system which does not allow them as the strict Kaynean antisymmetry proposal. We’ll have to motivate multiple functional heads into whose specifier positions the series of movements can take place. Following Jayaseelan (1999) these functional heads could be a series of IP internal focus positions.

Multiple Wh-

The next series of movements occurs into focus positions in the COMP,  positions which are well motivated in Rizzi (1992): Fine Structure of the Left Periphery.

Multiple Wh-

So here we have Focus-to-Focus movement of the wh-phrases, again resulting in the desired word order.

          Jayaseelan (2001) argues convincingly for movement of wh-phrases in Malayalam from a base generated vP internal position to focus phrases just above vP. There is no further movement of the wh-phrases to focus positions in the COMP because for some reason these focus positions are not generated in the COMP in these kinds of languages.

          But if there is only one series of movements our analysis predicts the following order for the wh-phrases within the IP:

Multiple Wh-

(10)    aar∂            aarkk∂       ent∂      koTuttu ?               (Malayalam)

          who.nom    who.dat      what     give-past

          ‘Who gave what to whom ?’

The system proposed by Jayaseelan (2001) includes a Topic phrase above the Focus Phrases and going by the idea that subjects are topicalized within sentences, we could think of the subject wh-phrase as being topicalized from the set of wh-phrases in focus positions

Multiple Wh-

This gives us the desired order S  IO  DO  V in these languages. Even in languages like Bulgarian with 2 series of nested movements, there can still be a topicalization of the subject wh-phrase but its effect is not visible in the word order

Multiple Wh-


Chomsky, Noam. 1973. Conditions on transformations, in A. Akmajian, (ed.), A festschrift for Morris Halle, Rinehart and Wilson, New York.

Chomsky, Noam. 2001.Derivation by phase. In M. Kenstowicz, (ed.), Ken Hale: A life in language, MIT Press, Cambridge Mass

Jayaseelan, K.A. 2001. IP-internal topic and focus phrases.  Studia Linguistica. 55.1 39-75.

Kayne, Richard. 1994. The Antisymmetry of Syntax. MIT Press, Mass.

Rizzi, Luigi. 1997. The fine structure of the left periphery. In L. Haegeman, (ed.), Elements of Grammar. Handbook of Generative Syntax. Kluwer, Dordrecht

Richards, Norvin. 1997. What moves Where When in Which language? Ph.D. Dissertation, MIT.

Wakanabe, Akira. 1992.  Subjacency and S-structure Movement of Wh- in situ.  Journal of East Asian Linguistics 1.  255-291.




K. G. Vijayakrishnan

0. Introduction

In this paper, I describe the realization of the different tonal contours attested in the standard, Majhi dialect of Punjabi. As tones are inextricably linked to stress in the language, I begin with a cursory look at the stress facts of Punjabi (Section 1)2A. Before proceeding to describe tones in Punjabi, I establish the pitch correlates of stress in words without any tonal specification drawing on work in progress by Balusu et al which examines the phonetic correlates of stress in five languages spoken in India. Having established the neutral, word level pitch melody for Punjabi, I briefly survey the literature on Punjabi tone, give a plausible account of tonogenesis in the language and arrive at a reasonably satisfactory transcription of the  two tonal contours attested in the language in 3. The main section of the paper (4.) presents a phonological analysis of tone in the context of the neutral, word level pitch melody in Punjabi. It is argued that this leads to the most explanatory account of the birth of two contour tones namely, a fall and a complex contour tone of a fall-rise in the language. It is argued that one of the major advantages of this approach is its account of tonal alternation attested in the language namely, the falling tone becoming a falling-rising tone in certain derived environments.

1. Word level stress in Punjabi

Unlike most languages spoken in India where word stress is a phonetic phenomenon not easily accessible for introspection, word stress in Punjabi is robustly perceptible and can be independently motivated elsewhere in the phonology of the language. For instance, it is the stressed syllable which is the domain of tone (i.e., the stressed                                                               hl    h        hl   h

syllable initiates the pitch movement) e.g., [mu:] ‘mouth’ but [mu'a:υre:] ‘proverbs’ (Sadanand and Vijayakrishnan 1993). Similarly, in echo word formation in Punjabi, it is the stressed vowel which is replaced in the echo reduplicated form. For example,

     hl  h                            hl  h   hl  h

['mi:na:] ‘month’ ~   ['mi:na: mu:na:] ‘month, etc’ (5 vi of Jha et al 1997) but [b∂gi:ca:]  ‘garden’ ~ [b∂gi:ca: b∂gu:ca:] ‘garden etc’ (15 iii of Vijayakrishnan 1998). Finally, the fact that stress is robustly phonological is proved by the shifting of


stress in words related through derivations e.g., [sε:] ‘bear (v)’ (base input /sah/)

h    hl  h                              hl                                                 h  hl(h)

but  [sa'ha:ra:] ‘support’ and [ kε:] ‘say’ (base input /kah/) but [ku'a:]  ‘make someone say’ (1 a8 ~b8 and a5 ~b5 respectively of Sadanand and Vijayakrishnan (1993).

Having established the phonological nature of word stress in Punjabi, we now turn to give a brief account of  stress placement in non-compounded words in the language. Comparing derived words with inflected words e.g., [la:ηa:] ‘to take off’ but [lu'a:] ‘cause to take’ and  [sa'la:(h)] ‘suggestion and [ sa'la:υa:~] ‘suggestions’[3A], though the final syllable in both the infinitival and the plural forms is heavy, the suffixes are not stressed whereas the heavy causative suffix is[4A]. However, not all derivational suffixes affect stress placement vis a vis the stress pattern of the base e.g., mitt∂r] ‘friend’ and its derivative [mit∂rta:] ‘friendship’ (Eliezer 1995). Suffice it to say for present purposes, while some derivational suffixes affect stress placement, other derivational suffixes and all inflectional suffixes are uniformly stress neutral. The former can be said to precede stress placement in a derivational model of phonology or in non-derivational parlance, we could stipulate that the [base + suffix]PrWd combine constitute a prosodic word encoded as the sub-categorization property of the suffix in question. As for the latter suffixes, they could be deemed to apply after stress rules have applied or, non-derivationally, they can be sub-categorized to be aligned to the prosodic word.

Turning to the details of stress placement in Punjabi, it must be noted that the language has a three way syllable weight distinction as in languages like Estonian (Prince 1980), Saami (Bye 1997), Arabic (Hayes 1995) and Hindi (Pandey 1989, Hayes 1991, 1995, Prince and Smolensky 1993). It has monomoraic light syllables (L), bimoraic heavy syllables (H) and trimoraic super-heavy syllables (S) which have a long vowel and a coda or a short vowel followed by two coda consonants. We assume- as is the standard practice in the literature-  that syllable bimoraicity is a universal upper limit and that surface trimoraic syllables should be analysed as virtual disyllables as shown in (1)- an adaptation of Bye’s compound analysis (1997) (Modified version of (10 b) of Bye 1997, p69) - a foot made up of a heavy and a light syllable.

1)      The representation of the super-heavy syllable

Super-heavy syllable

The stress facts of Punjabi are illustrated in the representative data in (2) below (part of 4) of Vijayakrishnan 1999):

2)      Word stress in Punjabi disyllables

i.                    Type I                   'LH   

'l∂hu:           ‘blood’                 'hirn    ‘deer’

ii.                  Type II                  L'S   

gi'la:f           ‘pillowcase’ g∂τcc       ‘drenched’  

Type II        I        H'S   

iii.      ja:'su:s                   ‘spy’           ta:'ri:f      ‘praise’

iv.      Type IV       'HH   

'Sa:di:          ‘wedding’    'ti:r∂t            ‘pilgrimage’

v.       Type V        'SH   

'cu:cca:        ‘chicken’     'ca:dd∂r   ‘sheet’

vi.      Type VI       'SS   

'ko:tυa:l        ‘magistrate   'go:rmint ‘govt.’

Examining the disyllables in (2 i-vi), we find that final heavy syllables are never stressed.[5A] Excluding words with super-heavy syllables, disyllables have a clear trochaic pattern of prominence of stressed followed by an unstressed syllables in a foot. If a word has only one super-heavy syllable, then it is stressed (with a trochaic foot built on the two virtual syllables) and if it has two super-heavy syllables, then the leftmost super-heavy syllable receives primary prominence. Assuming that the other super-heavy syllable also has a degree of prominence, we could say that words are fully parsed into disyllabic feet and that the leftmost foot is the head of the prosodic word.

Turning now to trisyllables, consider the data in 3) below (part of 4) of Vijayakrishnan (1999).

3)      Word stress in Punjabi trisyllables[6A]

i.        Type VII     H'HH

ka:'sa:g∂r     ‘potter’                 so:'ha:ga:      ‘leveller’

ii        Type VIII    L'HH

c∂'ritt∂r       ‘character’   s∂'re:ni:        ‘class’

iii.      Type IX       'HLH

'chicch∂ra:    ‘fat of meat’ 'g∂ηp∂ti:      ‘Lord Ganesh’

iv.      Type X        L'SH 

p∂'ti:lla:        ‘pan’           k∂'s∩TTi:     ‘touchstone’ (∩ stands for broken O)

v.       Type XI       H'SH

d har'me:nd∂r         ‘a name’

vi.      Type XII     'HHS

pa:kista:n   ‘Pakistan    'agg∂rυa:l     ‘a name’

vii.     Type XIII    ÈLHS

          'gulista:n     ‘garden’

viii.    Type XIV    'HLS

Once again we find that heavy syllables when final fail to receive prominence. However, unlike disyllables where a sequence of a light and a heavy syllable can make up a foot (2 i), in trisyllables, such feet are not constructed finally. For example, in (3 iii), avoiding the construction of such a foot, the foot is constructed on the first and second syllable, excluding the final syllable from the domain of the foot. But this restriction on avoiding an (LH) foot does not hold for non-final feet (3 vii) (assuming that a foot has been built on the final super-heavy syllable indicating secondary prominence).

Unlike real disyllables which are exhaustively parsed into feet, trisyllables, in principle, cannot be exhaustively parsed because of the universal prohibition on non-branching feet (McCarthy and Prince 1993). Now it is clear why the sequence LH was not footed in (3 iii)  but fully parsed in (3 vii). The pressure to parse exhaustively is evident in the latter creating a disharmonic trochaic foot. In the former, anyway one syllable can never be parsed (because of the foot binarity requirement) and when the choice between creating (HL)H and H(LH) comes up, the choice is weighted in favour of the more harmonic trochaic foot (HL).

Having established the systematic nature of word stress in Punjabi, we now turn to examine the pitch melody of words without any lexical tone and determine the pitch correlate of primary prominence in Punjabi.  

2. Pitch contour of words without tones

In this section, I report the relevant part of Balusu et al (work in progress) which examines some of the salient phonetic correlates of primary prominence in languages as varied as Punjabi (Indo-Aryan), Telugu (Dravidian) and Oriya, Bangla and Assamese (three related languages of the Eastern Magadhan Group). For all the languages examined, we restricted the data to di and trisyllables to analyze the correlate of initial, medial and final (if at all) prominence when these words were pronounced by three/four native speakers in carrier sentences in the sentence medial position. We examined three of six iterations of each word and computed significance for vowel length, syllable length, pitch levels and pitch movement.

As for Punjabi, the selected speakers three in all were uniformly speakers of the standard, Majhi dialect. We selected 23 tokens in all, carefully avoiding words with inherent tone and words with aspirates and sounds like [s] which are likely to induce perturbation of pitch as far as possible.[7A] Of the 23 tokens, six words exemplifying initial, medial and final stress (two each) are given below in  (4).

4)      Punjabi words examined for phonetic correlates of prominence

    Initial Prominence                    Medial Prominence                Final Prominence

i. 'ka:na:                         i. p∂Ta:ka:                     i. dar'ba:r

ii.'kanak                         ii. c∂m'ka:TTa:                ii.jaspa:l

Consider the pitch contours of words with initial stress in Figures 1. and 2. Ignoring the dip in pitch at the left edge (which may be attributed to either an intonational effect i.e., the rightmost ‘h’ of the preceding word and/or due to consonantal perturbation, we find that the stressed, initial syllable has the lowest pitch level in the word. There is a rise in pitch in the following syllable. The pitch level differences between the two syllables was found to be significant.

Figure 1

Figure 2

If we examine Figures 3. and 4. below, it becomes clear that once again, the stressed second syllable has the lowest, level pitch in the word and the syllable to its right has a high pitch once again as in disyllables with initial stress. We assume, as we did earlier, that here too the drop in the pitch in the first syllable is either an intonational effect and/or due to the perturbation caused by the initial consonant.

Figure 3.

Figure 4.

Consider finally the case of words with final stress in Figures 5. and 6. below. In words with final stress, we find that the final, stressed syllable has the lowest pitch, consistent with the generalization that the stressed syllable has a low pitch be it intial, medial or final in a word. However, unlike initially and medially stressed words, there is no high pitch or sharp rise in the final syllable in words with final stress. As in medially stressed words, here too there is a fall in pitch in the first syllable which we attribute to the final high of the preceding word.

Figure 5.

Figure 6.

To summarize, at the left edge of all the words, irrespective of the placement of prominence, we find that the initial consonant (part of the vowel has a drop form mid/high level. As we said earlier, it could be attributed to two sources. Firstly, it could be the transition from the high of the preceding word (an intonational effect) and/or, secondly, we can attribute some of the irregularities to the initial consonant. Therefore ignoring the left edge effect, if we examine the pitch contour of the stressed- unstressed sequence in both initially and medially stressed words, we find that uniformly stressed syllables have a significantly lower pitch than the following unstressed syllable.  And in words with final stress, the pitch remains low till the end of the word.

It could be conjectured that the word pitch melody is the sequence ‘lh’ with the ‘l’ associating with the stressed syllable and the ‘h’ associating with the following syllable, if there is one. The low on the stressed syllable may spread leftwards to lower the pitch of the preceding syllable, if there is one. The schema in (5) captures the pitch correlate of prominence on words without inherent tone.

5)      Pitch melody of words without inherent tone

Pitch Melody

Unlike languages like English and Japanese (Liberman 1975, McCawley 1968, Chew 1973, Haraguchi 1977, 1991, Poser 1984, Beckman 1986, Beckman and Pierrehumbert 1988, Kubozono 1988 (1993)) which select a ‘h’ on the stressed syllable, Punjabi consistently links a low tone to the stressed syllable in words without any inherent tone and high tone to the syllable on the right , if there is one.[8A] From now on, in our discussion, we ignore the pitch associated with the left edge of the word as it may be a spill-over from the final ‘h’ of the preceding word thereby becoming an intonational effect and/or it could be a perturbation associated with the initial consonant. Thus, we end up with the generalization that the word generates the ‘lh’ tonal melody of which the ‘l’ is associated with the stressed syllable. The rules of association given below account for the phonological pitch patterns of all the words without inherent tone.

6)      Pitch association rules for Punjabi                     (Preliminary version)

i.        The tonal sequence ‘lh’ is generated for every lexical category.

ii.       The tone ‘l’ is associated with the stressed syllable.

iii.      ‘l’ spreads to the syllable on the left of the stressed syllable, if present.

iv.      ‘h’ is linked to the syllable on the right of the stressed syllable, if present.

With this background pertaining to pitch patterns of words in Punjabi, we turn to the issue of contrastive tone in the language in the following section.

3. Tones and Tonogenesis in Punjabi

The literature on Punjabi, (Gill and Gleason (1969), Joshi (1989), Eliezer (1984), Sadanand and Vijayakrishnan (1993, 1998) and Sandhu (1968)) is consistent in maintaining that there is a three-way contrast in tones in lexical categories solely distinguished by tones as in (7) below from Sadanand and Vijayakrishnan (1998).

7)      Tonal contrasts in Punjabi

      l                            hl                                             hlh

i.  k∂l ‘machine’    ii. k∂l ‘tomorrow/yesterday’      iii. k∂l ‘to send’

While 7 i) is considered ‘neutral’ or toneless (actually it has low pitch as the word level ‘lh’ melody is neutralized to ‘l’ since the ‘h’ has no syllable to link to)[9A], 7 ii) and iii) have a falling tone and a falling-rising tone respectively. Whereas Gill and Gleason err on the abstract side suggesting that the three tones are static mid, high and low, Sadanand and Vijayakrishnan (1993) miss the phonological wood for the phonetic trees with their eyes close to ‘real’ pitch perturbations[10A]. In this paper, I show that the surface, dynamic pitch contour contrasts can be captured as the consequence of embedding a phonological static tone in the word level pitch contour ‘lh’ independently established for the language. The figures (7, 8 and 9) below display the pitch contours of the words in (7) above.

Figures 7. (top) and 8. (bottom)

Figure 9.

Eliezer (1984) takes a phonological stand arguing that the surface pitch contours are the by-product of underlying aspirates that may or may not be realized on the surface. He argues that the ‘lh’ (our hlh) and ‘hl’ contours can and must be attributed to underlying voiced aspirates in the onset and voiced aspirates in the coda respectively. While the former undergo de-aspiration and de-voicing, the latter only undergo de-aspiration. The data below is compiled from Eliezer (1984).

 8)     Representation of Punjabi tones

          Underlying Representation         Surface Representation    Gloss

Voiced aspirates in onsets generating a ‘lh’ tone


i.        /bhe:D/                                      [pe: D]                                     ‘sheep’


ii.       /ghi:/                                [ki:]                                 ‘clarified butter’

                                                   lh     h

iii.      /gho:Ða:/                          [ko: Ða:]                          ‘horse’ [11A] (Ð stands for a retroflex flap.)


iv.      /nha:/                               [na:]                                ‘bathe’


v.       /lha:/                                [la:]                                 ‘stir’

Voiced aspirates in the coda generating a ‘hl’ tone


vi.      /majh/                              [maj]                               ‘buffalo’


vii.     /labh/                               [lab]                                ‘find’


viii.    /b∂nh/                             [b∂n]                              ‘blind’


ix.      /th∂mh/                            [th∂m]                                       ‘pillar’

Eliezer cites orthographic evidence in support of his claim. However, Sadanand and Vijayakrishnan (1998) show that though a high tone (i.e., ‘hl’) is fairly well correlated with /s/ and /h/ and aspirates, there is no strict one-to-one correspondence synchronically between ‘underlying aspirates’[12A] and any one tone. The arguments I summarize below are from Sadanand and Vijayakrishnan (1998).

Firstly, ‘hl’ may not be uniquely traced to the coda contrary to Eliezer (1984).

9)      hl and underlying aspirates

Coda                                        Onset

                                 hl l                                                   hl

i a.     /kohlu/(H) [ko:lu:]            ‘oil press’  i b.       /buha/ (H)  [bua:]   ‘door’[13A]

                           hl                                                     hl   l

ii a.    /muh/(H) [mu:]       ‘mouth’     ii b./l∂mha/ (H)  [l∂ma:]     ‘split second’

Whereas the words in (a) have an underlying aspirate in the coda, the words in (b) have an aspirate in the onset.

Secondly, the presence of an aspirate in the cognate does not necessarily correlate with any tone in Punjabi as the data in (10) show.

10)    Lack of tone in the presence of an aspirate in the cognate

i.        /ba:r∂h/ (H) [ba:ra:~] ‘twelve’    ii.       /te:r∂h/(H) [te:ra:~]  ‘thirteen’[14A]

Thirdly, ‘hl’ may be present in words with a voiceless aspirate in the onset.

11)    hl and an initial voiceless aspirate

               hl   l                                                         hl  l

i.        khεb∂Ð ‘petty quarrel’       ii.       ph∂υi: ‘rabid female jackal’

Finally, tone may be idiosyncratically associated with a lexical item which has no aspirates at all.

12)    Tones without any aspirates in the segmental representation

            hl                                                       hl  l

i.        g ∂l    ‘cheeks’                         ii.       bε:ra: ‘bearer’(English)

            hlh                                                              hl    h

iii.      ra:t     ‘night’                                      iv.      mo:TTi:‘fat (female)’[15A]

Thus we see that the surface tones are not always attributable to underlying aspirates. Therefore, synchronically, it is reasonable to specify tones as part of the underlying representation of lexical entries in Punjabi.  Sadanand and Vijayakrishnan (1998) point out that even granting that tones can be traced to underlying aspirates historically, the case of tonogenesis in Punjabi is very unusual for several reasons. Firstly, unlike true instances of tonogenesis (Edmondson and Gregersn (1993) where voiceless consonants generate a high tone and voiced consonants a low tone, in Punjabi, both falling and rising-falling tones are presumably the result of voiceless as well as voiced aspirates albeit in different positions with respect to the syllable. Secondly, in cases of tonogenesis,  there is de-voicing only in the domain of high tone e.g., in Yabem (Ross 1993) the first person singular prefix is [ka- in the high toned context and [ga] in the low toned context as in [ka-teNg] (high tone ) ‘I weep (realis)’ but [ga-deNg] (low tone ) ‘I move towards (realis)’.  In Punjabi, high tones are randomly associated even with voiced obstruents (12 above) without concomitant de-voicing. It may have been the case that the neat scenario presented in Eliezer (1984) was the starting point for the advent of tone in the language. But synchronically, Punjabi has become a true tone language in the sense that the presence of tone is unpredictable and deemed to be lexicalized. Thirdly, cases like (12 iv and fn 10 ) indicate that the use of tone in the language is spreading in ways that are not yet very clear.[16A] Finally, Sadanand and Vijayakrishnan (1998) cite a three-way pattern in Punjabi words which have an /s/ in the cognate form in Hindi-Urdu. The data in (13) below is compiled from (5, 6 and 7 of Sadanand and Vijayakrishnan 1998).

13)    Pattern A             Obligatory tone

          Punjabi                           Hindi-Urdu            Gloss


i.        ko:                                  ko:s                      ‘a unit of distance’


ii.       υ∂r                                 b∂r∂s                    ‘rain’

Pattern B             Obligatory toneless

iii.      cu:s                                cu:s                       ‘such’

iv.      pya:s                               pya:s                     ‘thirst’

Pattern C             Free Variation

          Variant 1 ~   Variant 2


v.       pi:s              pi:                pi:s                        ‘grind’


vi.      kisda:           kida:            kiska:                    ‘whose’

Tone is well correlated with the deletion of the segment /s/ in the cognate. But as we said earlier, no segment type is either a necessary or a sufficient condition for the presence of either of the pitch patterns in a lexical item in the synchronic variety of the language.

4. Embedding Tones in Pitch Contours

In this section, we make an attempt at integrating the lexical tonal information associated with a word with the general pitch pattern of lexical items in the language. Taking up the hl tone first, it could be argued that it is a case of pitch reversal. In other words, in specified lexical items the normal pitch melody i.e., lh could be reversed giving the pattern hl. Thus, take a word like /k∂l/ meaning ‘tomorrow’, we could say that the lexical entry contains the information ‘T1T2 →T2T1 ‘ or something to that effect. However, this solution is both descriptively and explanatorily inadequate. Firstly, what could be the cause for this reversal? From the tonogenetic point of view, if aspirates did trigger this tonal pattern, what is the rationale for the pitch reversal? There can be none. Secondly, Joshi’s (1989 p64) description maybe interpreted to mean that there is a neutralization of the two distinct tone contours ‘hl’ and ‘hlh’ observed generally in verbs with lexical tone exceptionlessly resolved in favour of the latter pattern in causatives (see figures 10, 11, 12 and 13 below.) This tonal alternation cannot be explained if the hl tone is taken to be a case of pitch reversal.

14)    Tonal alternation in causatives in Punjabi

i.lab             ii. laba:                            iii. paj                  iv. paja:

Figure 10.

Notice that the non-causative form /lab/ and the non-causative /paj∂/ have the tonal patterns hl and hlh respectively. However, their causative forms have the pattern hlh. This alternation cannot be explained if we treat the hl tone as an instance of pitch reversal.

Figure 11.

Figure 12.(top) and Figure 13.(bottom)

The analysis we would like to advance is the following. Idiosyncratically, lexical items which are specified for tone do the following two operations. They uniformly introduce a H tone to the left of the pitch melody lh[17A] giving us the pitch melody Hlh. The pattern Hlh merely adding another tone (indicated by capitalisation) to the last h thus creating the complex pattern HlH.  The tonal specification(s) must be obligatorily linked within the domain. So the three patterns in the increasing order of complexity are then lh<  Hlh < HlH with the additional proviso that the non-tonal h is blocked from being linked in the domain in which the tone has been lexically inserted. Now the origin of tonogenesis can be stated roughly as follows. Aspirates anywhere in the base induce a H tone. Because of the nature of the neutral pitch melody, the tone is always sequence initial. In addition to the inserted tone, voiced aspirates promote the final h melody to a tonal status and simultaneously undergo de-voicing. This makes sense tonogenetically. Voiced aspirates as aspirates introduce a H tone and as voiced obstruents promote the final h to a H tone getting devoiced simultaneously. Of course, synchronically, the entire mechanism has been subverted with lexical entries idiosyncratically specifying any of the patterns in spite of the absence of aspirates (voiced or voiceless) anywhere in the word e.g., /g∂l/ ‘cheeks’ and / baggi:/ ‘ four-wheeled carriage’ with a hl pattern and  / ra:t/ ‘night’ and /mo:TTi:/ ‘fat (female)’ with a hlh pattern.

Let us first take up the Hl tone contour for discussion.

15)    the Hl         tone in Punjabi

i. ba:rla:                 ii. sala:(h)               iii. sala:Va:~

Figure 14.

Figure 15.

Figure 16.

This is the pattern Hlh .Notice that the final h does not link to any vowel in Figure 14 and in the plural form (Figure 16)? A simple reformulation of the linking conventions in  (6) will account for this fact.. The reformulated version of (6) is given below as (15).

15)    Pitch association rules for Punjabi lexical entries                 (Revised version)

i.        The pitch sequence ‘lh’ is generated for every lexical category.

ii.       The lexically specified H tone is inserted to the left of the pitch melody.

iii.      The ‘l’ pitch is always associated with the stressed syllable.

iv.      The lexically specified tone is always linked to the stressed syllable.

v.       Spread pitch/tone to the left if a syllable is present.

vi.      If the stressed syllable is associated with a unique pitch specification, link the remaining pitch specification to the syllable on the right, if present in the pitch domain 

(15 vi) requires some explanation. The idea is that tonal contours are uniquely associated with the stressed syllable and that, normally, only the last tonal/pitch  specification on the stressed syllable can spread right as linking of a new specification  would create a second pitch contour within a single tone domain. This explains why the final pitch h does not associate with any tone bearing element in the figures in 14, 15 an 16[18A] as the association of the free h would create a second pitch contour from l to h.  As opposed to this state of affairs, when morphology creates an additional tone-stress domain, generally shifting the stress to the right, then the ‘free’ h does link within the new domain creating a hlh pattern as can be seen the figures 10 and 11 above. And this tonal alternation is not restricted to the causatives as the examples in (16) below from Sadanand and Vijayakrishnan (1993) illustrate.

16)    Tonal alternation in derived contexts.

           Hl                                     h Hl   h

i.        mu:              ‘mouth’       mu'aυre:              ‘proverb’

           Hl                                            h  Hl h        Hl  h

ii.       kε:               ‘speak’                 ka'haηi:/'ka: ηi:/   ‘story’

Now we turn to the tonal contour HlH generally associated with underlying voiced aspirates which, we argue, promote the final h pitch to a tonal status.

17)     The HlH tone in Punjabi

i. kabra:                 ii. kabra:haT       iii. ane:ri:

Figure 17.

Figure 18.

Figure 19.

Compare the association of HlH in monosyllables with that of words which have an extra syllable to the right of the stressed syllable. It is clear that in all cases, the lexically specified tones must associate within the tone domain. However, whereas all three pitch specifications must obligatorily associate with the same syllable in monosyllables and final stressed syllables as tones must be obligatorily linked, in words which have an extra syllable to the right of the stressed syllable, the final H tone links not to the stressed syllable but to the final syllable (this is very clearly illustrated in /kabra:aT/ and /ane:ri:/ - figures 18 and 19). We can now revise further the association statements in (15) as (17) below.

17)    Pitch association rules for Punjabi lexical entries                 (Final version)

i.        The pitch sequence ‘lh’ is generated for every lexical category.

ii.       The lexical specifications *i , *j

a.       insert a H tone to the left of the pitch melody (*and *j )

b.       promote the final h to a H tone (*j )[19A]

iii.      The ‘l’ pitch is always associated with the stressed syllable.

iv.      Lexically specified tone(s) must be linked.

v.       The leftmost tone is always linked to the stressed syllable.

vi.      Avoid triple linking, if possible.

vii.     Spread pitch/tone to the left if a syllable is present.

viii.    The rightmost pitch specification is linked to the syllable to the right of the stressed syllable iff

a.       the stressed syllable is linked to a single pitch specification.


          b.       a new pitch domain has been created by stress sensitive suffixation.

The lexical specifications *i  << *j   are in the increasing order of markedness as the former requires one tonal specification resulting in a falling tone and the latter two specifications resulting in a falling –rising tone (17 ii a and b). It is better for a fall-rise to be realized on two consecutive syllables constituting the stress foot if possible. However, if another syllable is not available to the right of the stressed syllable, a fall-rise on one syllable is tolerated. Finally, the final h pitch (not tone) is realized iff the stressed syllable is uniquely linked to an l pitch within the base domain (viii a). This ensures distinctiveness of surface pitch melodies vis a vis Hl and HlH tones in the language. The scenario in (viii b) is possible only when a stress sensitive suffix is added to the base.

18)    Linking of the final h pitch in an extended tonal domain

Extended Tonal Domain

When stress sensitive suffixation takes place (viii b), the tonal distinctness between Hl and HlH is neutralized although the rise on the final syllable is created by the final, linked h pitch and not a lexical tone. A small wrinkle needs to be ironed out at this point. It will seem as though (18) will apply only to alternations like /mu:/ ‘mouth ~ /mu'a:υre:/ ‘proverb’ and /kε:/ ‘speak’ ~ /ka'ha:ηi:/ ‘story’ where there is an extra syllable to the right of the stressed syllable. What about the causatives where the Hl ~ Hlh alternation is attested without exception? Traditionally, in the literature, the causative morpheme has been transcribed as –a:] i.e., a bimoraic morpheme. We had also stated earlier that this was the only bimoraic final syllable which attracts stress in Punjabi. Neither stem final long vowels nor long vowels of other suffixes attract stress. The hypothesis I would like to advance is that the causative morpheme is really not bimoraic but trimoraic with an extra long vowel. The durational evidence that I furnish is given below in Table 1 where I show that the duration of final /a:/ in non-causative disyllables is significantly shorter than that of the causative morpheme with or without tone.

19) Evidence for representing the causative morpheme as –a:a]causative

% duration



Non-causative disyllables


i. ka:na:




ii. rasa:


iii. ta:ra:


iv. ba:rla:


v. va:sta:


vi. vatna:


Causatives disyllables


i. kara:[20A]





Ii. laba:


Ii. paja:


The causative /a:/ is significantly, distinctly longer than the non-causative /a:/. We may therefore give it a trimoraic representation making it a virtual disyllable. If that is the case, we do not have to make an exception of the causative morpheme. It attracts stress because it is super-heavy and it allows the association of an extra pitch precisely because it is trimoraic- a virtual disyllable.[21A]  

5. Conclusion

I have tried to show in this paper the why and the how of tonogenesis in Punjabi. High tone must have been triggered by aspirates and /s/ and /h/. Of these sounds, /s/ and /h/ always delete after inducing the high tone. And the marked fall-rise contour was probably triggered by voiced aspirates which lost their voicing as a result of inducing a second H tone. If that were the end of the story, Punjabi would have merely acquired a two way, static, tonal discrimination namely, low i.e., without tonal specification and a newly created high tone. But the creation of the complex tonal contours   Hl and HlH was the result of embedding new lexical information in the word level pitch melody lh commonly found in Punjabi (and many other languages spoken in India), probably a historical relic from the Sanskrit rule of pitch association (see footnote 8). The basic pitch association rule linking a low pitch to the stressed syllable has stood its ground resulting in the creation of tonal contours.

Well, if tonal contours had to be created and only two were needed, why not stop with just Hl and Lh? The answer is, once again, quite simple. The latter could not have been a new tonal contour as it would have been indistinguishable from the basic lh pitch melody already existing in the language. Therefore, the only way to create a further distinct tonal contour was to combine a fall and a rise in the same domain. Finally, the hypothetical question remains: Punjabi has created a Hlh melody by adding a tone to the left. Why did it not create a symmetrical lhL with a tone on the right? Once again the answer lies in the history of the language. Since tonogenesis was initiated by aspirates and /s/ and /h/, only a high tone could have been induced. Moreover, a low pitch was already well entrenched, linked to the stressed syllable and plain voiced consonants which have a tendency to induce low pitch had no role in tonogenesis in the language. These are the advantages of accounting for the tone patterns in the synchronic variety of the language by embedding them in the normal pitch melody associated with words without inherent tone. Finally, the claim that tone is embedded in the word pitch melody finds support in instances of tone alternation where morphology creates an extended domain for pitch to link and the unlinked final h pitch of the word pitch melody gets linked in all such cases resulting in an apparent change of tonal pattern from Hl to a HlH. 


Balusu, R, Mahanta, S, Mohanty, R and Vijayakrishnan, K.G. (in progress). The low rise on stress in Punjabi, Telugu, Oriya, Bangla and Assamese. Ms. CIEFL.

Beckman, M. 1986. Stress and non-stress accent. Netherlands phonetic archives 7. Foris: Dordrecht.

Beckman, M and Pierrehumbert, J. 1986. Intonational structure in English and Japanese. Phonology 3.

Bye, P. 1997. ‘Representing ‘overlength’’ in Booij, G et al (eds). Phonology in progress- progress in phonology: HIL phonology papers III. Holland Academic Graphics: The Hague.

Chew, J.J. 1973. A transformational analysis of modern colloquial Japanese. Mouton: The Hague.

Edmondson, J.A and Gregorson, K.J. 1993. Western Cham as a register language, in Edmondson, J.A and Gregorson, K.J. (eds). Tonality in Austronesian languages. University of Hawaii Press: Honolulu.

Eliezer, J. 1984. Some aspects of the autosegmental phonology of English and Punjabi. Unpublished MPhil dissertation. CIEFL: Hyderabad.

Eliezer, J. 1995. The logical structure of phonological representstion. Unpublished PhD dissertation. CIEFL: Hyderabad.

Gill, H.S and Gleason, H.A. 1969. A reference grammar of Punjabi.Punjabi University: Patiala.

Haraguchi, S. 1977. The tone pattern of Japanese: an autosegmental theory of tonology. Kaitakusha Publishers: Tokyo.

Haraguchi, S. 1991. A theory of stress and accent.Foris: Dordrecht.

Hayes, B. 1985. Iambic and trochaic rhythm in stress rules. Berkeley Linguistic Society 11.

Hayes, B. 1995. Metrical stress theory: principles and case studies.Chicago University Press: Chicago.

Joshi, S.S. 1989. The phonology of the Punjabi verb: a polysystemic analysis. Classical Publishing Company: New Delhi.

Kubozomo, H. 1988 [1993]. The organization of Japanese prosody. PhD dissertation. University of Edinburgh. Kurioso Publishers: Tokyo.

Liberman, M.Y. 19875. The intonational system of English. PhD dissertation. MIT.

McCarthy, J.J. and Prince, A. 1993. Prosodic morphology I: constraint interaction and satisfaction. Ms. University of Massachusetts, Amherst and Rutgers University.

McCawley, J.D. The phonological component of a grammar of Japanese. Mouton: The Hague.

Pandey, P. 1989. Stress in Hindi. Lingua.

Poser, W.1984. The phonetics and phonology of tone and intonation in Japanese. PhD dissertation. MIT.

Prince, A. 1980. A metrical theory for Estonian quantity. Linguistic Inquiry 11.

Prince, A and Smolensky, P. 1993. Optimality theory: constraint interaction in generative grammar. Ms. Rutgers University and University of Colorado.

Sadanand, K and Vijayakrshnan, K.G. 1993. ‘Pitch patterns associated with the glottal fricative in Punjabi: A preliminary investigation’, in Rama Rao, C et al (eds). Osmania Papers in Linguistics; 19. Osmania University: Hyderabad.

Sadanand, K and Vijayakrishnan, K.G. 1998. ‘High tone in Punjabi: A case of deaspiration and tonogenesis’. CIEFL Occasional Papers in Linguistics; 7. CIEFL: Hyderabad.

Sandhu, B.S. 1968. The tonal system of the Punjabi language. Parkh: Punjab.

Vijayakrishnan, K.G. 1998 [2000]. Conflict resolution: The case of Anti-faithfulness in reduplicative morphology. Paper presented at the First Asian GLOW: CIEFL. CIEFL Occasional Papers in Linguistics 8.

Vijayakrishnan, K.G. 1999. The disyllabic trochee in Punjabi. Paper presented at the International Conference on Stress and Rhythm, December 1999, CIEFL, Hyderabad and SALA 2001, Konstanz, Germany.



 1A I am very grateful to Kamlesh Sadanand for making my work on Punjabi possible. This paper is an extension of the ideas in earlier papers jointly undertaken by both of us. Her well discriminated data, excellent sense of perception and challenging new ideas which we frequently, heatedly debated have contributed to this work as well. Of course, I am solely responsible for errors of analysis if any. I would also like to thank Rahul Balusu for his enormous patience in helping me with the PRAAT platform.

[2A] For a fuller, Optimality Theoretic account of Punjabi stress see Vijayakrishnan 1999.

[3A] We have not included tonal information here as our attention is on the placement of stress in these forms.

[4A] See section 4. for a discussion of the representation of the causative morpheme

[5A] The only systematic exception to this generalization is the causative suffix.

[6A] The last three types are due to Kamlesh Sadanand (p.c). And the fourth is, perhaps, a possibility. It would be possible on the surface only if the first syllable is a closed syllable, otherwise, the vowel in the medial light syllable might get deleted, merging this type with a disyllabic 'SS. In the standard dialect, the voiced aspirate in 3 v) devoices and deaspirates generating a tone.

[7A] See section 3. below for tonal effects of underlying aspirates, [s] and [h] in Punjabi.

[8A] This is reminiscent of the Sanskrit rule associating a high pitch to the right of the accented syllable (Kiparsky pc). Curiously, in all the five languages we examined, we found the word level pitch melody to be ‘lh’ with the rule associating a low tone to the stressed syllable and a high tone to the adjacent syllable/mora with minor language–specific variations.

[9A] Joshi calls it PC2 quite infelicitously, implying as though such words have a specific lexical tone which is not the case at all.

[10A] Another problem with Sadanand and Vijayakrishnan is that since the words were pronounced in isolation, there was a list intonation effect with a final fall on most words leading to a wrong generalization about the over all pitch contour associated with the word.

[11A] Eliezer’s transcription is also not quite accurate. For instance, what he transcribes as ‘lh’ is actually ‘hlh’ in monosyllables and ‘hl’ on the stressed syllable and ‘h’ on the following syllable if present.  See discussion and pitch graphs in the following section.

[12A] A cover term for aspirates, /s/ and /h/.

[13A] ‘H’ stands for Hindi-Urdu.

[14A] This generalization is true for all the Hindi-Urdu cardinals between twelve and eighteen.

[15A] This may not hold for all speakers of the standard dialect. Also, the related word /mo:TTa:/ ‘fat (male)’ has no tone (Kamlesh Sadanand pc). Joshi (1989) too notes that for his wife some words have acquired a ‘hl’ tone which are ‘toneless’ for him (see fn 9, p 73).

[16A] It would be interesting to undertake dialect-specific surveys involving large number of speakers and a fairly large fixed common vocabulary to examine the tendencies pertaining to tone at work in the language. Needless to say, such a massive task requires funding (totally lacking in the Indian context).

[17A] Actually, the direction of the insertion of an h tone need not be specified at all as it follows from a general principle namely, OCP.

[18A] We assume that stress neutral affixes are outside the domain of tones as well. Therefore, only those pitch levels which are already linked are available for further linking at this stage. Notice that the l of sala:(h) is available for the plural suffix (figure 16).

[19A] ...and de-voice an obstruent, if present.

[20A] Contrary to popular belief, there is no significant extra length in the causative morpheme caused by the falling-rising tone.

[21A] The fact that the final h does not link to stressed, final super-heavies which are also virtual disyllables is problematic. Perhaps, the difference between final super-heavy virtual disyllables and causative super-heavy syllables is that only the latter have a  tone bearing element – a vowel. More work would have to be done in the area of consonantal perturbation and pitch and tones in Punjabi before a fully fleshed out analysis is possible. 




Shakuntala Mahanta

0 . Introduction

In this paper I analyze the prosodic phonology of the Negative Phrase (henceforth NEG P) in Assamese, with special attention to double negative constructions. Apart from prosodic movement, there are also instances of optionality in NEG phrasing in Assamese. I analyze this optionality as focus driven negation since the NEG clitic in the language is inherently a focus marker. Even though the semantics of NEG Phrasing ordinarily can be explained by a syntax phonology mapping, there are instances of NEG P which defies a syntactic explanation. To offer a unified account of the NEG P paradigms in the language is the goal of this paper.

In my analysis, Optimality theory takes care of the alignment of NEG phrases with phonological phrases by the constraint ALIGN NEG R PhP L. Moreover, while simple NEG P can be explained by such alignment constraints, instances of over generation of the NEG P which poses serious problems for a syntactic/semantic analysis of the phenomenon under consideration, can be explained in terms of OT by simply ranking Dep NEG P as the lowest constraint.

          The paper is organized as follows: The various motivations for the prosodic phonology of the NEG marker in Assamese (§ 1), double negation and so called triple negation (§2). A discussion on Focus, Phrasing and intonation in Assamese (§ 3).  Focus and the negative clitic (§ 4). An Optimality theoretic account of negation in Assamese (§5).

1.       Prosodic phonology of the NEG marker in Assamese

There are quite a few motivations to argue for a prosodic phonology of the NEG marker in Assamese. The motivations come from the general behaviour of  the NEG element in all positions of a sentence. The position of the NEG element relative to the verbal element is always governed by well defined prosodic rules.

 NEG Procliticization

NEG  [NV ]- in Assamese is a proclitic which is left-adjoined to the verb from which it copies the vowel.

1)i.     kha     ‘eat’~  na'k ha       ‘neg-eat’      ii. khεl                   ‘play’ ~  nεkhεl ‘’

That the NEG element in the words given above is not an independent morpheme is clear from the following facts:

·        The NEG element NV does not form a prosodic word (henceforth Pwd) by itself and therefore, it is phonologically parasitic on the verbal argument.

·        The copied, underlined vowel (in the examples in 1i) of the NEG element is much shorter than the vowel of the base.

·        The verbal base retains its prominence unlike ‘normal’ prefixation which shifts prominence leftwards. (Mahanta 2001).


2)i.     'hadharn    ‘ordinary’ (∩ stands for broken O)   ~ '∩hadharn                  ‘extraordinary’

·        The NEG clitic initiates its own focus contour L*, apart from the contour of the focused element  - the verb.

The Prosodic Constraint on NEG Movement

When NEG is not a proclitic (i.e. it meets the prosodic word requirement) it can occur on either side of the verb. While the canonical position is the right of the verb, it can also shift to the left of the of the verb under emphasis.

4.       mi    'bhat  khua   nai

          I        rice    eat     neg

          ‘I have’nt eaten rice’

5.       m∩ i    bhat    '[nai  khua]

          ‘I have’nt eaten rice’

Even this liberty is restricted to those NEG markers, which equals the prosodic word requirement. Once this requirement is exceeded the NEG marker cannot move to the left of the verb, unlike the NEG marker which equals a Pwd.

6.       mi    'bhat  khua   'na-si-lu

          I        rice    eat     neg- past

          ‘I hadn’t eaten rice.’

7. *    mi    'bhat  'na-si-lu      khua

·        NEG can be optionally moved (under emphasis) to the left of the verb making the NEG+verb combine a compound prosodic word e.g., 5. But if the base of NEG is larger than a prosodic word movement is blocked e.g., 7.

At this point it is interesting to compare the Assamese NEG marker’s behaviour with Bangla, a language related to Assamese typologically as it belongs to the same group of languages (Eastern Magadhan subgroup of Indo Aryan languages). The Bangla NEG marker in the same contexts2 is fixed to the right side of the VP and never moves from there irrespective of the size and shape of the NEG marker (na and nei being the commonly used forms).

8.       Assamese                       Bangla                                      Gloss

a.       mi          na  khau               ami        khabo   na           I won’t eat

          I         neg     eat               I          eat   neg

b.       mi          khua      nai                 ami        khaii     nei                 I havn’t eaten

          I        eat      neg             I       eat     neg

c.       mi                    naikhua             *ami  nei  khaii               I havn’t eaten

2.       The Prosodic phonology of double negatives in Assamese

9.       mi    bhat    kh ai    nth - ka     nhi

          I        rice    eat     neg-stay-prt neg-be-agr

          ‘It’s not that I’m staying without eating rice.’

NEG Movement under Emphasis

10.     mi    bhat    na [kh ai th - ka]    nhi

I        rice   neg-eat            stay-prt     neg-be-agr

          ‘It’s not that I’m staying without eating rice.’               

The sentences 9 and 10 have the same interpretation. Under emphasis, the inner neg can be left-adjoined to the main verb and the complex predicate forms a P phrase with the NEG clitic.

NEG Copy Under Emphasis : Triple Negation

11.     mi    bhat    na kh ai         nth - ka    nhi

          I  rice  neg-eat-agr  neg-stay-prt neg-be-agr

          ‘It’s not that I’m not eating rice’

The sentences 9 , 10 and 11 have the same interpretation. Though there are three overt NEG markers in 11, the sentence is interpreted as an affirmative sentence exactly like 9 and 10 with just two overt NEG markers. Therefore, it is clear that the copied   (reduplicated) NEG to the left of the main verb is not accessible for interpretation at LF. This construction can also be interpreted as doubly emphasized double negation in order to assert affirmation. The so called triple negation occurs under the idiosyncratic focus of the main verb khai. Moreover, these sentences are not marked in the language. Examples of these sentences are abundantly available:

12.     mi nki                nuzua              nhi

          I        neg say(agr) neg go(agr)  neg be-agr

‘It’s not that I went without saying’             

13.     m∩i          namari      nεpεlua                 nh∩i

          I        neg beat(agr)         neg throw(agr)                 neg be-agr

          ‘It’s not that I have not beaten (him) to death’

Such examples with verbal compounds are quite productive in the language.

For the syntax of negation this is an overgeneration of the NEG element which cannot be mapped to its affirmative meaning. I tackle this apparently problematic construction with the help of focus and I follow Lahiri and Fitzpatrick Cole (1999)  in specifying lexical tone to the Negative focus signifier of Assamese. Intonational information is decomposed and stored in the lexicon in the form of tunes . These sentences can be interpreted as the NEG clitic specified in the lexicon with its focus tune, which can be used to indicate assertion under extraordinary circumstances.

 Let us deal with some more issues bearing on our discussion of intonation, phrasing and focus, before trying to arrive at an explanation for double negatives. 

3.       Intonation3 , Focus and  Phrasing in Assamese

Phrasing and rhythmic grouping in Assamese

There has been no detailed investigation of intonation in Assamese4. Since the area of my investigation is restricted in the sense that intonational evidence is sought to validate whatever is phonologically quite palpable, I analyze the rudimentary intonational framework of a few sentences only. That Phonological phrases (henceforth P phrases) are the domain of a lot of prosodic effects in Assamese will be borne out in the discussion in this paper. Before proceeding to analyze the interaction of focus and phrasing triggered by the Negative clitic, let us consider the phrasing of neutral affirmative sentences given below:

14.     (mi)                  (bhat  k au)))P

I(nom)                   rice    eat

          ‘I eat rice’

15.     mi           ( bhat) P  (k ai  thakU)P

          I (nom)                    rice   eat     stay

          ‘I keep eating rice’

The phrasing in 14 and 15 indicate that the rhythmic requirement is an important prerequisite of phrasing in Assamese. In the sentences above, while the subject forms one phonological phrase, the complex predicate forms another group.  Otherwise, the verb adjoins to whatever is available on its left. 5

 Lahiri and Hayes (1981), has shown that Bengali respects phonological phrase boundaries in the following positions:

17      a   P ( V

b.     P ( X1, X2 …. X3)P

c.     Focus)P

Our preliminary investigations have shown that Assamese does not always follow a similar pattern. As exemplified in 9, the final verb  forms P phrase with the object, and complex predicates form one P phrase. Lahiri and Hayes (1991), Lahiri and Fitzpatrick Cole (1999) has given numerous segmental evidence to justify their claim. We will not go into the segmental and assimilatory details that they have furnished as evidence, but we will generally agree that the Assamese phrasing boundaries are located immediately before a verb in a complex predicate. 

          As already stated, there has been no detailed investigation of intonation in Assamese, so we cannot say anything definitive about nuclear declarative accents. A study on focus revealed that the focused material is associated with a low pitch accent6 at the left edge and a high boundary tone. We can draw a parallel with Bangla focus where the low pitch accent of the focus contour L*H is left aligned to the initial P word of the focused phrase and the right edge is characterized by a high boundary tone.

3.1     NEG procliticization and the phonological phrase.

The position of clitic plus host groups in the prosodic hierarchy has received considerable attention from phonologists, without arriving at a consensus. I will briefly summarize a few proposals.

a)     as a separate level of the prosodic hierarchy, located between the prosodic word and the phonological phrase( Nespor and Vogel, 1986)

b)    as simply a variety of  pword   (Selkirk,1986)     

c)     as pwords in some languages and phonological or intonational phrases in other languages (Inkelas 1989, Zec and Inkelas, 1991)

d)    Selkirk (1995) has shown that it is not necessary to have an independent prosodic category for host and clitic groups. She argues that such sequences can be of four different types. The function word and lexical word may be separate p words:

In this case the function word is not a clitic. The function word does not form a P word by itself in other cases.

φ ((Fnc ) ω(lexical wd))   free clitic

φ(ω (Fncω( lexical wd))        affixal clitic

φ(ω (Fnc  lexical wd))    internal clitic   

(φ-  phonological phrase           ω -prosodic word)

Even though nothing crucially hangs on which clitic group the NEG P in Assamese belongs to, we assume from the discussion above that the NEG clitic and the verb coalesce to form a phonological phrase in Assamese. Functional words, which are not P words, combine with lexical words to form phonological phrases. The NEG proclitic does not form a prosodic word independently and therefore, it is always left aligned to a VP, and forms a phonological phrase.

 It has been noticed that that clitic group formation rule does not respect syntactic constituency (Nespor and Vogel, 1986). Klavans (1982,1985) observed that a clitic can attach syntactically in one direction and phonologically in another direction. The nonisomorphicity of the Assamese NEG clitic is also demonstrative of a similar trait, showing that the clitic group formation in Assamese is in agreement with the behaviour of clitic groups in other languages as well.  

3.2 Complex predicates

 Lahiri and Fitzpatrick Cole (1999) give the following account for complex predicates in Bengali.

In complex predicates the second stem is always a verb  …. The first stem can be either a verb, which is always a perfective, or a noun which is always unmarked. For the verb- verb predicates it is more difficult to get a reading where each separate verb literal meaning is preserved. For the complex predicate meaning the words must be phrased together in a single P phrase. Complex predicate, like simple verbs, phrase separately from the preceding material.

V – V Complex predicates                           verb – verb reading

          Sue porA    ‘to lie down’                             sleep PERF fall –VN

          mere phe|A        ‘to kill’                           beat PERF throw –VN

The citation form of the verbs is the so-called verbal noun form in –A, which is usually translated as an infinitive.

Assamese complex predicates phrase similarly under normal circumstances. An example has already been given in 5. Let us repeat the same example for the sake of illustration

19.     mi           ( bhat) P  (k ai  thakU)P

          I(nom)                   rice      eat  stay

          ‘I keep eating rice’

20.     Here are some more instances of Assamese V –V complex predicates

V – V Complex predicates                           verb – verb reading

 bhri p∩ra                           ‘to be full’ full PERF fall -VN

bur        zua                               ‘to drown’ sink PERF go- VN

  However, the NEG clitic disturbs the normal phrasing of complex predicates. This will be elaborated in the following section.

4. Focus and the Negative clitic

Insitu NEG in Assamese is also inherently indicative of focus. This is apparent from the tune text alignment of  the NEG marker in preverbal positions. When the NEG marker shifts from the right of the verbal element to its left there is also a shift in the tone. In the sentences in 21 and 22, the NEG shifts to the left to indicate emphasis.  Whereas in 19 the pitch accent is borne by '(bhat) to indicate that it is rice which is emphasized, in 20 the locus of emphasis shifts to the preverbal position. Consequently, the focused verb as well as the clitic fall under the intonational rubric of the focus contour.  The optionality in phrasing is focus driven as the change in P-Phrase break in 22 shows us. Corresponding to the NEG movement and the alignment of the NEG element to the P phrase domain of the verb, what is interesting is the tune which shifts along with the shift of the NEG element.

21.     HEAD                   NUCLEUS                    

          L*      HP      L*                                                    LI

          ((mi            )P   (bhat) )P       (khua nai)P      )I

          I                  rice              eat     neg

          ‘I have’nt eaten rice’


22.     HEAD                                     FOCUS               

          L*        HP                         L*         LI                  


          ((mi bhat)P                  '(nai khua)P                   )I

I        rice    neg              eat

          ‘I haven't eaten rice’

The tune text alignment in 21 and 22 above, shows the pattern of emphatic negation only in sentences  where the NEG element constitutes a prosodic word. Notice that it was earlier stated that L*H indicates focus in Assamese. But the focus contour of the sentence in 20 above indicates that L* is enough to indicate focus. It will be seen that the negatives, under special circumstances, initiate their own focus tune apart from the focus represented on its phonological host. This will be borne out even by the focus in double negatives.

          The syntactic composition of the double negative constructions is also important in this discussion. Notice that there are three verbal elements in these constructions, where the final copula obligatorily takes needs the NEG marker, whereas the complex predicate can have NEG on either / or both  the verbal components. There is a conflict between the demands of the focus bearing NEG clitic which prefers to host on the nearest verb, and the main verb which vies for focus.

   The Assamese NEG clitic when it is used for focus not only lexically selects the focus element; it also shows its inherent focus specificity. Under acute emphasis, the focus specified NEG clitic aligns to the left of the main verb, and thereby provides focus information. Consequently, even though prominence is retained on the phonological host which bears the pitch accent, the NEG clitic specifies additional focus information. The focus contour of the double NEG sentences given below will systematically show that L* is the lexical tone of the NEG clitic for emphasis.

23.     HEAD                    NUCLEUS

          L*HP                       L*H                             LI

          ((mi)P      (kh ai  nth - ka)P (nhi)P    )I

          I                 eat     neg-stay-prt neg-be-agr

          ‘It’s not that I’m staying without eating .’

Focus in this sentence follows the normal focusing pattern illustrated earlier i.e., L*H. Normally, the NEG clitic hosts on the nearest verbal element, which is the light verb thka in this case.  The NEG clitic normally aligned to the second part of the compound predicate, includes the main verb also in the focus contour L*. Further, this pattern can also be interpreted as the main verb’s need for focus and therefore the association of L* with the whole P phrase. Unlike Lahiri  and Cole, we do not consider the phrasing (kh ai n'th - ka)P ( where the clitic adjoins between two verbal compounds) problematic. We had earlier shown  that Assamese like Bangla,  prefers to phrase complex predicates as one unit, even though the  NEG clitic  adjoins to the second verbal element . We can also exercise the option of phrasing the main verb kh ai  and the NEG + light verb  n th - ka separately. 7

24.     ((mi)P(bhat)P      (kh ai)P ( nth - ka)P         (nhi)P    )I

 We consider the P phrasing given above invalid because we do not have the problem of associating the focus contour with the concerned clitic. As we have already said, in this construction NEG is already specified with focus information, whereas, the main verb also needs to be focused. Consequently, the L* means focus on the entire P phrase (kh ai n'th - ka)P , whereas the H is the phrasal  boundary tone.

25      HEAD                           FOCUS      

           L*HP                  L *H I                            LI

          (mi)P       (na kh ai th - ka)P   (nhi)P     )I

I                 neg-eat            stay-prt     neg-be-agr

          ‘It’s not that I’m staying without eating .’                     

 26.    HEAD         FOCUS          FOCUS   

          L*HP          L*H                L*H                        LI

          (mi)P       (na kh ai)P     (nth - ka)P (nhi)P    )I

          I  rice  neg-eat-agr  neg-stay-prt neg-be-agr

          ‘It’s not that I’m not eating’

The intonational contours of the double negatives given above shows unprecedented patterns, even though they can be broadly classified as L*H. As we now know, both the NEG clitic and the main verb need to specify focus information. In 23, NEG had aligned to the nearest verb available from the right. In 25, there is NEG movement and as a result, the NEG which had earlier adjoined to the nearest verbal element, adjoins to the main verb kh ai. This indicates that the main verb kh ai  is also competing for focus information. This results in the idiosyncratic focus of the main verb, where the NEG clitic like the full NEG in 20 and 21 introduces its lexically specified tone and adds a prefix to the pitch accent of the main verb, this prefix interacts with the focus contour of the argument and the interaction results in the L*H contour .

          Therefore, like the Bangla focus clitic which provides intonationl information in the form of the H* pitch accent, the Assamese Negative clitic is also lexically specified with the tone L*. The complex tune –text alignment of these sentences (25 & 26), when compared to the tune of the sentence in 23, in the diagrams below, will bear testimony to our contention.

          The tune in 26 is however even more complex, even though the NEG phrase can be roughly called a L*H. We assume that there is double focusing along with NEG copying, which results in the downstepping of the second NEG phrase. The problem that also needs to be addressed is the P Phrasing in 26  which does not conform to our predicted phrasal boundary. Assuming that Assamese also shares Bangla Pphrasing, it was shown that complex predicates normally do not break up into two phrases. The P Phrasing in 24 shows that all the NEG bearing verbal elements have phrased separately. This weakens our earlier postulation that complex predicates phrase separately, but as it has already been shown in section 2 that Assamese considers rhythmic grouping an important criteria of phrasing. When uttered, the sentence forms three separate groups with equal prominence on all the component negative phrases. This also explains the apparent nonavailability of the meaning of this sentence at LF. The semantically empty clitic is copied and reduplicated to enhance rhythmic prominence ( like echo word formation which also functions like an intensifier ), and the lexical tone associated with the negative focus is aligned to it to augment the intensification. The L* of the negative clitic interacts with the pitch accent of the main verb and produces the complex L*H.

          Finally, the boundary tone Lmay also be interpreted as the tone of the final NEG P . Even we do not know anything about the boundary tones of declarative phrases in Assamese, we can still say that the NEG clitic in the final phrase interacts with the boundary tone to produce a low boundary tone.

More evidence for lexically specified L* on NEG

          Research has shown that tone in languages may not be a parameter, but more in the nature of a continuum; such that there may be languages which generally do not use tone as a contrastive feature, but may have tones on certain lexical items along with their inventory of intonational morphemes. The analysis of the Assamese NEG clitic has been on these lines and more evidence comes from the behaviour of the verb zanu.

27.     zanu                             nazanu

‘know’                           ‘neg know’  (first person, simple present)

(I ) know                         (I) do not know

Very often, the NEG clitic  za  is dropped and replaced   by a  marked low tone. The pictures below show the difference between      zanu(aff)     zanu(neg).

zanu (aff)


The fact that Negation can be expressed by the tone despite the non occurring NEG may point in the direction of an inherent low tone in the NEG marker.

5.       An Optimality theory account of   Negative Focus in Assamese

I want to show with the help of Optimality Theory (Prince and Smolensky, 1993, that the NEG paradigms in Assamese can be explained without much difficulty.

Earlier analyses of syntax – phonology interface( Truckenbrodt 1995, 1999, Selkirk 1984      used constraints like these:

28.     ALIGN XP, R: The right edge of a syntactic XP is aligned with the right edge of a P phrase. This constraint ensures that all syntactic elements are parsed into p phrases.

An extension of the theory of edge alignment motivated by Truckenbrodt (1995, 1999) states that the effects of edge alignment may be suppressed in certain positions by WRAP XP.

29.     WRAP XP: Each XP is contained in a phonological phrase.

WRAP XP affects phrasing by favouring larger phrases over smaller ones. Thus ALIGN XP, R ensures a stricter parsing, WRAP XP prefers wrapping up all constituent phrases into a single P Phrase.

These constraints are unsuitable for our analysis as the prerequisite of these constraints is that Phonological phrases meet the requirement of syntactic phrases. P phrases taken up for consideration in this paper fall short of this requirement.

I postulate the following constraints in order to account for the Focus driven P phrasing of  the negative sentences in Assamese.

We will need a constraint to account for the NEG marker at the left edge of the VP.

30.  ALIGN (NEG, R, P Phrase, L):    Align NEG to the left of the rightmost phonological phrase.

The constraint which ensures that the focus is on the verb is:

 31.  MV – FOCUS                :         Focus the main verb in NEG V phrases.

To prevent insertion of NEG elements on all the verbal components, we have the constraint below:

32.  DEP (NEG)                      :         Do not insert a NEG clitic

Before we find out how the constraints are ranked respective to each other, we will have to find another constraint which will take care of the phrasing of the sentences. I postulate the constraint PARSE to account for the phonological phrasing of the NEG phrases.

33. PARSE PHRASE                      :         Phrases should be constituents of P phrases.

This constraint makes it obligatory for all phrases to be parsed into P phrases. However it is not binding on this constraint to parse all XP’s into separate phrases.

Furthermore, it is quie essential to parse NEG phrases into separate phrases. Therefore, I have the constraint given below:

34.     NEG P                                     :         Parse NEG phrases.

This constraint stipulates that NEG Phrass should be part of phonological phrases. This constraint is violated by any NEG P which is not a phonological phrase by itself.

 Given these constraints, it will be seen how the optimal candidate is selected by the interaction of these constraints. PARSE will be ranked as the highest constraint as it is important to parse phrases, keeping in mind the rhythmic profile of the language under consideration. Also ALIGN NEG and VERB FOCUS are important in the formation of NEG phrases. So all these three constraints are ranked as below so as to generate well-formed NEG phrases of the language.






[[[ khai nothoka]nohoi]





ALIGN     NEG                 R, PhP,L

?1. (khai nothoka)P (nohoi)P





2.  (khai)P (nothoka)P( nohoi)P




3. (khai)P (nothoka nohoi)P


 * *!



4. (khai nothoka nohoi)P




5. khai (nothoka)P( nohoi)P





  As the tableau shows, it is essential to form P phrases with all the terminal elements and at the same time it is mandatory for the final NEG phrase to form a phonological phrase. Notice that in the optimal candidate the need to focus the main verb is satisfied despite the fact that the NEG clitic is not adjoined to the main verb because the compound predicate (the main verb and the light verb with which the clitic is adjoined ) forms a P phrase and therefore the verb also falls under the focus contour . ALIGN NEG R, PhP L is lowest in this hierarchy, as the optimal candidate itself incurs one violation of this constraint.

Therefore, all the candidates incur violation of this constraint except candidate 4, where all the phrases are wrapped into one. Candidate 3 violates all the constraints except PARSE, whereas 4  fatally violates only this constraint, and therefore it is discarded from the process of evaluation. 5 violates NEG P as well as the low ranked ALIGN because of the unparsed main verb khai. Thus the harmonic candidate is selected as it parses all the component phrases as well the final NEG Phrase and it also focuses the main verb. It makes one violation of ALIGN NEG, which is affordable as NEG is lowest ranked in the hierarchy.

[[[na khai thoka]nohoi]





ALIGN     NEG                 R, PhP,L

?1. (nakhai thoka)P (nohoi)P





2. (nakhai)P (thoka)P( nohoi)P





3. (nakhai)P (thoka nohoi)P





4. (nakhai thoka nohoi)P



5.  nakhai (thoka)P( nohoi)P





The same ranking schemata given in 35 takes care of the interaction of the various constraints resulting in the desired output. First of all, PARSE ensures that all terminal elements are part of PhP s and NEG P parses all the NEG Phrases. MV FOCUS being the next evaluator imposes the restriction of  focusing the main verb. Candidate 2  violates the constraint ALIGN  NEG R, PhP, L and therefore differs from the harmonic candidate  in the number of violations of  ALIGN  NEG  R, PhP,L. 4 is not optimal because it violates PARSE NEG PHRASE,  which is ranked highest in the hierarchy. Candidate 5 incurs the violation of the highest constraint PARSE along with ALIGN  NEG R, PhP, L and MV FOCUS, and therefore eliminated from the evaluation procedure

  However, at this stage, we will have to introduce the constraint DEP NEG. This constraint was vacuously satisfied in the other NEG constructions exempified till now


  [[[ khai] nothoka]nohoi]





ALIGN     NEG                 R, PhP,L


?1.(nakhai)P (nothoka)P (nohoi)P






2. (na khai)(nothoka nohoi)P





3. (nakhai nothoka)P   ( nohoi)P


*! *!




4. (nakhai nothoka nohoi)P






5.  na khai (nothoka nohoi)P






6. .( khai)P (nothoka)P (nohoi)P



All the candidates in 38 except 6, violate DEP NEG P. It is the lowest constraint in the hierarchy; therefore, even though the optimal candidate incurs one violation of DEP NEG P, it is selected.  Candidate 2 violates the inviolable NEG P twice. 5 violates all the three highly ranked candidates as a result of not phrasing  na khai .6 does not violate DEP NEG, but that does no help as it does not have main verb focus, and fatally violates ALIGN NEG R, PhP,as well. 3 and 4 club NEG P’s together violating NEG P.

          Therefore, the final ranking schemata for NEG P ‘s in Assamese is given below:







          DEP NEG P

In this ranking schemata it is obligatory for syntactic phrases to form part of P phrases

At the same time NEG phrases should always form Pphrases on their own. These two undominated constraints coupled with Main Verb Focus constructs NEG P in such a manner that three different types of phrasing are possible to express focused Negation in Assamese.


1. I am grateful to Prof. K.G. Vijayakrishnan who lets us run away with pieces of his ideas (including this). However, the usual disclaimers apply. 

2. Bangla also has instances where the NEG element shifts to the right of the verb, but it does not shift in these contexts. It will be interesting to compare the Neg element in the two languages extensively, but it is not within the scope of the present paper.

3. We assume the intonational theory of Beckman and Pierrehumbert(1986, 1988), Hirschberg and Pierreheumber (1996), which formulate the compositional theory of tones. Static tones (L,H) independent of  linguistic text can be associated to phrasal structure to make up tune menings. In this paper the slightly modified  association conventions introduced by Hayes and Lahiri(1991) is used.(T* to stressed syllable; TP to an intermediate = phonological  phrasal boundary; TI to an intonational boundary.

4.  Even though I use the standard intonational conventions of dividing the IP into a nucleus (containing the accent and suffixes) and intonational heads, it is too early to designate specific meanings to tones. Theefore I have simply associated the different parts of the IP to its components without  signifying meanings( other than FOCUS ) .

5. The phrasing shown in neutral declarative sentences is tentative and needs a lot more investigation before they can be established as the rules of phrasing in the language. Preliminary studies showed these patterns, it is open for future research to find out the validity of the phrasing here.

6. Pitch  accent is on the strongest syllable of  the P word . In Assamese, word level prominence is on the leftmost syllable of the sentence, unless there is a following heavy syllable. (Mahanta,2001)

7. The recursive option is considered unavailable, following the debate over recursivity.


Beckman, Mary and Janet Pierrehumbert. 1986. Intonational structure in Japanese and English. Phonology Yearbook 3: 255–309.

Balusu, Rahul, Caroline Fery and Shakuntala Mahanta. 2002. Focus and Intonation in a few Indian languages ( Forthcoming)

Fitzpatrick-Cole, Jennifer. 1994. The Prosodic Domain Hierarchy in Reduplication.Unpublished Ph.D. dissertation, Stanford University.

Fitzpatrick-Cole, Jennifer. 1996. Reduplication meets the phonological phrase inBengali. The Linguistic Review 13: 305–356.

Hayes, Bruce and Aditi Lahiri. 1991. Bengali intonational phonology. NaturalLanguage and Linguistic Theory 9: 47–96.

Inkelas, Sharon. 1989. Prosodic Constituency in the Lexicon. Ph.D. dissertation,Stanford University. Published by Garland.

Inkelas, Sharon and Draga Zec. 1988. Serbo-Croatian pitch accent. Language 64:2,227–248.

Lahiri, Aditi and Jennifer Fitzpatric- Cole 1999. Emphatic Clitics and

Focus Intonation in Bengali. In René Kager & Wim Zonneveld (eds.) (1999) Phrasal Phonology. Nijmegen: University of Nijmegen Press. 119-144.

Inkelas, Sharon and Draga Zec (eds.). 1990. The Phonology-Syntax Connection.

Chicago: Chicago University Press.

Jayaseelan, K,A,. 2000. IP Internal Topic and Focus Phrases. CIEFL Occasional Papers in Linguistics.

Klavans J. 1982. Some Problems in a Theory of Clitics. Ph.D Dissertation: University College London, 1980. ULC.

Klavans,J. 1985. The Independence of Syntax and Phonology in Cliticization. Language 61, 95-120

Mahanta,Shakuntala. 2001. Some Aspects of prominence in Assamese and Assamese English. M. Phil Dissertation, Central Institute of English and Foreign Languages, Hyderabad.

McCarthy, John & Alan Prince 1999. "Faithfulness and Identity in Prosodic

Morphology". The Prosody Morphology Interface ed. by René Kager, Harryvan der Hulst & Wim Zonneveld, 218-309. Cambridge: Cambridge UniversityPress.

Nespor, Marina and Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris.

Pierrehumbert, Janet. 1980. The Phonology and Phonetics of English Intonation.Ph.D. dissertation, MIT.

Pierrehumbert, Janet and Julia Hirschberg. 1990. The meaning of intonational contours in the interpretation of discourse. In Philip R. Cohen, Jerry Morgan and Martha E. Pollack (eds.) Intentions and Communication. Cambridge, MA: MIT Press. 271–311.

Prince, Alan & Paul Smolensky 1993. "Optimality Theory: Constraint Interaction in Generative Grammar". Ms., Rutgers University and University of Colorado,Boulder.

Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. PhonologyYearbook 3: 371–405.

Selkirk, Elisabeth O. 1984. Phonology and Syntax: The Relation between Sound and Structure. Cambridge: MIT Press.

Truckenbrodt, Hubert 1995. Phonological Phrases: Their Relation to Syntax, Focus,and Prominence. PhD thesis, Massachusetts Institute of Technology.

Truckenbrodt, Hubert 1999. "On the Relation between Syntactic Phrases and Phonological Phrases". Linguistic Inquiry 30:2.219-255.






The term phonological awareness has initiated a lot of interest among researchers who are interested in exploring the intricate relationship between reading development and phonological awareness. Research shows that in English, syllabic awareness precedes phonemic awareness and among the intrasyllabic units rime precedes onset. The present study tests the above hypothesis with Bangla speaking adult illiterates. The subjects were given an oddity task where they were asked to listen to triplets and say the odd word. The results show that in majority of the responses subjects said the odd word on the basis of onset and following vowel, which goes against the rhyme hypothesis. This is perhaps because the orthography of Bangla is different from English in the way phonemes are mapped to graphemes, and literacy might have played a role in determining the responses of the subjects.


Goswami and Bryant (1990) have shown that the developmental levels of phonological awareness appear to emerge sequentially. The awareness of syllables, onsets and rimes appear to emerge at the ages of 3 and 4, long before most children go to school. The awareness of phonemes usually emerges at the ages of 5 and 6, when children have been taught to read for about a year. Therefore we may say that the awareness of onset and rime appears to be a precursor of reading, whereas the awareness of phonemes develops as reading is taught.

There are very few studies, with the exception of Morais, Alegria, and Content (1987), (who worked with child illiterates), that have selected adult illiterates as subjects and tried to find out their phonological awareness. Studies of phonological awareness have been done mostly with children (pre-school and in-school) and have tried to show how a child’s phonological knowledge helps him or her to learn to read.

In the following section we will briefly discuss the literature on phonological awareness among pre-school children, and how the concepts of onsets, rimes and analogies play a significant role in the development of children’s reading. Later, in this study, we will try to establish the connection (if any) between the phonological awareness of pre-school children and adult illiterates.

Levels of phonological awareness among children

Phonological skills in pre-school children can be measured at a number of levels. There are at least two levels of phonological awareness that might be important in explaining the link with reading (Goswami, 1991). The two levels are the phonemic level and the syllabic level.

The Phoneme

In English, phonemes usually correspond to individual letters or pairs/trigrams of letters in printed words. This relation is called the grapheme-phoneme correspondence. For example, in the word cat, the phonemic transcription of the word is /æ/. Here the letter ‘c’ stands for the sound /k/, the letter ‘a’ stands for the sound /æ/, and the letter stands for the sound /t/.

The Syllable

A syllable consists of speech sounds: vowel and consonants. If a syllable consists of just one sound, it is usually a vowel. If a syllable has more than one speech sound, one of them is a vowel and the rest are consonants. At the intrasyllabic level, single-syllable words are broken up into the linguistic units of onset and rime. The onset consists of the initial consonant(s) of a word (for example, in the word ‘tap’, ‘t’ is the onset; in the word ‘plea’, ‘pl’ is the onset). The rime corresponds to the vowel(s) and the following consonant(s) of a syllable (for example, in the single-syllable word ‘tap’, ‘ap’ is the rime, and in ‘plea’ ‘ea’ is the rime.

The distinction between the phonemic and the intrasyllabic level is an important one because studies (Morais, Alegria, and Content, 1987; Goswami & Bryant, 1990; Goswami, 1999) have shown that the phonemic awareness develops only once children are taught to read or receive specific training about phonemes i.e., phonemic awareness is a consequence rather than a precursor of reading. What develops early is the ‘gross phonological units: syllables, onsets and rimes (Goswami & Bryant, 1990).

As far as the early availability of onsets and rimes is concerned, studies (Treiman & Zukowski, 1996; Kirtley, Bryant, McLean & Bradley, 1989) have compared the development of phonological awareness of onsets, rimes and phonemes in school children by using different phonological awareness tasks.

Phonological awareness tasks

Treiman and Zukowski (1996) used a same-different judgement task based for the beginning or end sounds of words. In the beginning sound task, the words either began with the same onset, as in plead and play, or shared only the initial phoneme, as in peak and pray. In the end sound task, the words either shared the entire rime, as in peak and beak, or shared only the final phoneme, as in peak and trick. The results of the study showed that 4-5 year old children found the onset-rime version of the same-different task much easier than the version based on phonemes. Only 6-year olds, who had been reading for about a year, performed equally well in both versions of the task.

Kirtley, Bryant, McLean & Bradley, (1989) showed that the development of onset-rime awareness precedes an awareness of phonemes that are not onsets. Children had to select the odd word at the onset-rime level as well as the phonemic level. For example, in word triples like top, rail, and hop, the odd word was to be selected on the basis of the whole rime. In triples like mop, lead and whip, the odd word was to be selected on the basis of the final phoneme. The results showed that 4-5 & 6 year old children showed a selective deficit in the phoneme version of the task while they were quite at ease with the onset-rime version of the task.

Rimes before Onsets

Goswami (1991, 1998) while endorsing the findings, adds that young children make more analogies in reading for the shared letter sequences at the ends of words rather than at the beginning. (In other words, the notion of ‘rime’ develops in children before the notion of ‘onset’). For example, more analogies are made by children between ball and call, than between camp and caps, because ball and call share a spelling sequence that reflects the rime, whereas camps and caps share a spellingsequence that reflects the onset and part of the rime. This finding therefore involve predictions about rhyme: a child who uses ball to read call is making the prediction that the two words will have rhyming pronunciations and children who are good at rhyming make more such analogies (Goswami, 1990). The finding also predicts that a word sequence like call and ball will facilitate reading better than a word sequence like camps and caps.

Thus the special relationship between awareness of rhyme and the development of children’s reading seems well established, at least for English. A question arises, however, about the relationship between phonemic awareness and reading instruction. Since phonemic awareness develops after about a year of reading instruction, a plausible hypothesis is that reading instruction is a causative factor in the development of phonemic awareness, i.e., the developmental sequence of rime-onset-phonemic awareness is not purely autonomous, but is facilitated by reading instruction. If so, the investigation of the phonological awareness of adult illiterates assumes significance. As mentioned earlier, not much literature is available on if adult illiterates show phonological awareness at the intrasyllabic level.

Aim of the study

In the present study we are interested in finding out the kind of phonological awareness that develops in Bangla speaking adult illiterates. In other words, we shall investigate whether it is intrasyllabic awareness or phonemic awareness that emerges in Bangla speaking adult illiterates.

A few words on the orthographic-syllabic-phonemic mapping in Bangla

The orthography of Bangla is different from the orthography of English in the way the phonemes are mapped to graphemes. If we look at a single-syllable word in Bangla (with the structure of CVC), we will see that the graphic unit (the medial vowel) falls into the preceding consonant and constitutes a single unit and this particular unit is neither rime nor onset. Therefore the mapping from phonemic object to orthography in Bangla destroys the syllabic units of onsets and rimes.

Thus we can logically infer that in the case of Bangla speaking children, rime awareness does not seem to play a role in facilitating reading.

Research design and procedure

The experimental material

The experimental material consisted of a set of 27 words. The words were commonly used and real words. These 27 words were arranged in 9 triplets. As shown in the following table, two words out of three in a row shared the same rime, e.g., ka:m, na:m, and ka:n (in the first row). Again, two other words in the same triplet shared the same onset and vowel. For example, again in ka:m, na:m, and ka:n (in the first row), ka:m and ka:n share the same onset + vowel. In the table below, the letter (O) at the extreme left of the cells stands for ‘onset + vowel’, and the letter (R) at the extreme right of the cells stands for ‘rime’. (Thus the cells which are marked with (O) at their extreme left share an onset and the following vowel, while the cells which are marked with (R)at their extreme right share a rime).

Table 1
The set of stimuli given to the students.

(O)      Ka:m     (R)

               Na:m            (R)                          

(O)          Ka:n

(O)       Ba:l      (R)

                Ka:l             (R)

(O)          Ba:n

(O)       Ni:l       (R)

                Bi:l              (R)

(O)           Ni:r

(O)       No:l      (R)

                Bo:l             (R)

(O)           No:r

(O)     Dhu:p    (R)

                Cu:p          (R)

(O)           Dhu:s

(O)      Ka:n     (R)

                Pa:n            (R)

(O)           Ka:l

(O)      Ja:l       (R)

                Kha:l           (R)

(O)           Ja:m

(O)      Mol      (R)

                Khol           (R)

(O)           Mom

(O)      Broj      (R)

                Koj              (R)

(O)           Bron

The Task

The task was an oddity task. The subjects were to select the odd word.


Four male adult speakers of Bangla in the age group of 21-23 years took part in the study.
We need to mention here that a close look at the results shows that there is a pattern that emerges from the responses of the subjects. Subject S1 and S2 always made onset+vowel analogies in order to say the odd word from the triplets whereas, S3 almost always made rime analogies to say the odd word from the triplets. This seems to have a bearing on the literacy level of the subjects which is briefly discussed below.

They had been labourers in a construction company (Larsen&Tubro). Initially they told the researcher that they had not received any formal education, and they could not read or write in Bangla. But in informal interaction, it was found that three out of the four subjects (S-1, S-2, S-4) could read monosyllabic words like ‘ek’ (one), ‘boi’ (book) etc. and with the help of the researcher, the three could also collectively read the names of movies from a newspaper. They then told the researcher that they had been to a literacy camp in their village for a very brief period (a week or so). The significance of the interaction is that it tells us about the subjects’ rudimentary acquaintance with the Bangla orthography. But when it came to the reading of the words in the experimental material independently and individually and without the help of the researcher, they failed.

Presentation of the stimuli

The position of the odd word in each of the triplets was systemically varied to ensure that the subjects did not develop a pattern of guessing. In the following table (Table 2) we present the stimuli as given to the subjects.

Table 2

"S" stands for "same" and "D" stands for "different."

Ka:m (S)

Na:m (S)

Ka:n (D)

Ba:n (D)

Ba:l (S)

Ka:l (S)

Ni:l (S)

Bi:l (S)

Ni:r (D)

No:l (S)

No:r (D)

Bo:l (S)

Dhu:p (S)

Cru:p (S)

Dhu:s (D)

Ka:l (D)

Ka:n (S)

Pa:n (S)

Ja:l (S)

Kha:l (S)

Ja:m (D)

Mol (S)

Khol (S)

Mom (D)

Bron (D)

Broj (S)

Koj (S)

In the experimental session, the subjects were individually read out these words in triplets.

Results of the group

We need to mention here that a particular subject S4 was unable to understand the experimental task; therefore the data collected from S4 has not been included in the study. So the study is based on the data collected from three subjects (S1, S2, S3). Table 3 below gives us the complete results. The responses are classified according to whether the analogies are made by rime or by onset+ vowel.

Table 3

Conditions Analogies made
by Rime
Analogies made
by Onset+Vowel
Total number of
6 (out of 27) 21 (out of 27)

Out of 27 sets of triplets, rime analogies were made only for 6. This seems to go against the rhyme or rime hypothesis of Goswami et al.

Coming now to individual responses, in the following table (Table 4) we present the subjects’ responses in identifying the odd word.

Table 4

Odd Word

By ‘rime’

By ‘onset +vowel’

Set 1: ka:n in ka:m, na:m, and ka:n

Subject:    S3

Set 1: na:m in ka:m, na:m, and ka:n

Subject:          S1,  S2

Set 2: ba:n in ba:n, ba:l, and ka:l

Subject:      S2, S3

Set 2: ba:n in ba:n, ba:l and ka:l

Subject:          S1

Set 3: ni:r in ni:l, bi:l, and ni:r

Subject:      S3

Set 3: bi:l in ni:l, bi:l, and ni:r

Subject:           S1, S2

Set 4: no:r in no:l, bo:l and no:r

Subject:       S3

Set 4: bo:l in no:l, bo:l and no:r

Subject:         S1, S2

Set 5: dhu:s in dhu:p, cru:p, and dhu:s

Subject:       S3

Set 5: cru:p in dhu:p, cru:p, and dhu:s

Subject:       S1, S2

Set 6: ka:l in ka:l, ka:n, and pa:n

Subject:       S3

Set 6: pa:n in ka:l, ka:n, and pa:n

Subject:       S1, S2

Set 7: ja:m in ja:l, kha:l, and ja:m

Subject:       S3

Set 7: kha:l in ja:l, kha:l, and ja:m

Subject:       S1, S2

Set 8: mom in mol, khol, and mom


Set 8: khol in mol, khol, and mom

Subject:      S1, S2, S3  

Set 9: bron in bron, broj, and koj


Set 9: koj in bron, broj, and koj

Subject:      S1, S2, S3

If we closely look at the subjects’ responses in the table above, we will see that there is a consistent difference in the responses of the subjects. Except for the sets 2 and 9, Subject S3 always said the odd word on the basis of rime and Subjects S1 and S2 always said the odd word on the basis of onset+vowel.


Our study shows that the rhyme/rime hypothesis does not automatically extend to Bangla adult illiterates. A particularly intriguing aspect of our study is the grouping of our subjects in “rime” and “onset+vowel” analogy response-types.

We shall now speculate on two possible reasons for our results. As discussed earlier, the orthography of Bangla (or for that matter, any Indian language) is different from that of English in the way phonemes are mapped to graphemes. The mapping from phonemic/phonetic object to orthography in our languages destroys the syllabic units of onset and rime.

Given this, the literacy level of the subjects becomes an important factor in determining their responses. As discussed earlier, two subjects (S1, S2) who were considered illiterate always made onset +vowel analogies. Interestingly, these were the subjects who went to a literacy camp for a week and could read a few monosyllabic words with the help of the researcher. The logical conclusion we can draw from the finding is that the element of literacy might have destroyed the subjects’ initial syllabic awareness. In contrast, subject (S3) who was totally illiterate, made only rime analogies.

Alternatively, it may be the case that syllabic awareness is itself a cultural construct. English speaking children are much more exposed to nursery rhymes and that explains their inclination towards the rime. Whereas, in most Indian languages (for example, in languages like Tamil, Bangla etc.) alliteration is very much a prominent feature, especially in speech and early rhymes and that may well argue that our inclination towards the onset +vowel pattern is probably ingrained. (This explanation was given to me by Professor K. G. Vijaykrishnan)

Clearly, the present study is just a beginning. The exploration of phonological awareness in speakers of Indian languages is a necessity.


I would like to thank Prof. R. Amritavalli and Prof. K.G. Vijaykrishnan for their very kind help and valuable comments throughout the study.


Goswami, U. (1986). Children’s use of analogy in learning to read: developmental study. Journal of Experimental Child Psychology, 42, 73-78.

Goswami, U. (1990). A special link between rhyming skills and the use of orthographic analogies by beginning readers. Journal of Child Psychology and Psychiatry, 31, 301-311.

Goswami, U., & Bryant, P.E. (1990). Phonological Skills and Learning to Read. Hillsdale, NJ: Lawrence Earlbaum.

Goswami, U. (1991). Learning about spelling sequences: The roles of onsets and rimes in analogies in reading. Child Development, 62, 1110-1123.

Goswami, U. (1999). Integrating orthographic and phonological knowledge as reading develops: Onsets, rimes and analogies in children’s reading. In The Human Brain. MIT Press: Massachussets.

Kirtley,C., Bryant, P., Bradley,L. (1989). Rhyme, rime and the onset of reading. Journal of Experimental Child Psychology, 48, 224-245.

Liberman, I. Y., Fischer, F., Carter,B. (1974). Explicit syllable and phoneme segmentation in the young child. Journal of Experimental Child Psychology,18, 201-212.

Morais, J., Algeria, J., & Content, A. (1987). The relationship between segmental analysis and alphabetic literacy: An interactive view. Cahiers de Psychologie Cognitive, 7, 415-438.

Treiman, R. (1988). The internal structure of the syllable. In G. Carlson and M. Tanenhaus (Eds), Linguistic Structure in Language Processing (pp. 27-52). The Netherlands: Kluegar.

Treiman, R., and Zukowski, A. (1991). Levels of phonological awareness. In S. Brady and D. Shankwilier (Eds), Phonological Processes in Literacy. Hillsdale: Earlbaum.

Treiman, R., and Zukowski, A. (1996). Children’s sensitivity to syllables, onsets, rimes and phonemes. Journal of Experimental Child Psychology,61, 193-215.



HOME PAGE | BACK ISSUES | A Prolog Analyzer/Generator for Sanskrit Subanta Padas | From Spelling Bee to Misinterpretation of Telephone Conversations | Useful and Advanced Topics for Conversation in Intensive Courses: Heritage Learning and Current Social Learning | Understanding Proxemic Behavior | CIEFL Occasional Papers in Linguistics, Vol. 10 | The Two Lives of -unnu in Malayalam: a Response to Amritavali and Jayaseelan | The Order of the Inflectional Morphemes in Arabic | Serial Verbs with the Light Verbs ja: and de in Oriya | Multiple Wh-Fronting and Superiority: A Nested Movement Analysis ...
| Stress and Tone in Punjabi | The Prosodic Phonology of Negation in Assamese | Phonological Awareness in Adult Illiterates: Onsets, Rimes, and Analogies | CONTACT EDITOR

K. G. Vijayakrishnan, Ph.D.
Centre for Linguistics
Central Institute of English and Foreign Languages
Hyderabad - 500 007, India