On the formation of complex predicates in Bengali

1 Introduction

Verbal constructions that are generally described as Complex predicates are usually constituted of two items, one chosen among various categories of words: noun, verbal forms, adjective, preposition, adverb, onomatopoeic, etc., and the other, a normally inflected verb. The first item is called a Pole and the second one a Vector (a.k.a. explicator, operator or light verb) (cf. Hook 1974: 17; Masica 1976: 143; Dasgupta 1977: 68; Butt 2003). These predicates are generally put into two different categories on the basis of the syntactic category of the pole: (i) Compound verb (1, 2) that categorially involves a non-inflectional verbal form (e.g. a participle, an absolutive or a past gerund), and (ii) Conjunct verb (3, 4) that involves categories other than the verb. In other words, a ‘Compound verb’ results from the combination of two different verbs, and a ‘Conjunct verb’, from the combination of a verb and another word that belongs to some other category. The main characteristic of Compound and Conjunct verbs is that they must denote one single action. ^[1]

There seems to be no general consensus either about (i) the domain or about (ii) the process of formation of such predicates (cf. Butt 2003). In this article, we will try to contribute to these two problems in the light of Bengali data constituted of sentences that are acceptable according to the competence of the author in this language. ^[2] This article is divided into 6 sections. In Section 2, we briefly present a couple of solutions that have been proposed by authors who prefer to handle complex predicates in syntax, and then we argue why syntax is not the right place to account for them. In Section 3, after a brief description of W(hole) W(ord) M(orphology) → the model we shall use in this article – we show how complex predicates can be handled in morphology. In Section 4, we argue that complex predicates are words. In Section 5, we point out a few problems of handling complex predicates in morphology in the light of atomistic models of word formation, i.e. models that must involve units smaller than the word, before, finally, we draw conclusions in Section 6.

2 Bengali complex predicates: Are they formed in syntax?

In our view (inspired by Singh 1996) one must try to answer the following three questions of Domain, Status and Process in order to account for any linguistic construct.

1. In which domain are these constructs formed, for example, in Syntax or in Morphology?

2. What is the status of these constructs, should they be considered Phrases (or Clauses), or are they Words?

3. Through which particular process(es) are they formed?

The unmarked stand vis-a-vis the domain of the formation of complex predicates is that they are formed in syntax. Grimshaw and Mester (1988) describe Japanese complex predicates as phrases whereas Kageyama (2001) describes them as Word+ (word plus), an intermediate category between the word and the phrase. For Matsumoto (1996) Compound verbs are biclausal syntactic structures. ^[3] Authors who claim that words may be formed in syntax propose a couple of processes: (i) A(rgument) L(inking) P(rinciple) (cf. Lieber 1983; Lieber 1992, 2004 among others) ^[4] or (ii) A(bstract) I(ncorporation) (cf. Kageyama 1991), ^[5] both based on the semantic relation between the argument noun appearing as the pole and the verb appearing as the vector. According to Grimshaw and Mester (1988) and Matsumoto (1996) Japanese complex predicates are formed through syntactic processes which usually form phrases and clauses.

It is unclear how word-formation processes like ALP or AI can account for Bengali Compound verbs which are constituted of two different verbs (1, 2) or for Bengali Conjunct verbs in which the pole is represented by a word belonging to a category other than the noun (4) (because such poles cannot appear as the argument of a predicate). Unlike Compound verbs (1, 2), Conjunct verbs (3, 4) are not bi-clausal constructions and it is unclear how they can be handled in frameworks like Grimshaw and Mester (1988) or Matsumoto (1996). However, one can always argue that at least some Conjunct verbs, such as (3), can be formed through ALP or AI and some others with syntactic processes.

The question of status (ii) and that of process (iii) are both in fact redundant, in that it is the domain that automatically determines the status of the construct in question. For example, linguists generally do not disagree with the fact that syntactic constructs such as phrases and clauses are semantically more transparent and formally less cohesive as compared to words and idioms. If this is in fact the case, it would be reasonable to handle them in some domain other than the syntax. Processes, on the other hand, can determine neither the domain nor the status of the output because, logically, the same (type of) processes may output different things in different domains. For example, as we have seen above, linguists claim to account for the formation of words in syntax with processes that are otherwise used to form phrases and clauses.

In the following, we argue, on the basis of empirical data, why syntax is not the right domain for handling complex predicates in Bengali.

i. Unlike phrases, Complex predicates are generally semantically opaque. For example, the Compound verb likhe pheleche ‘he has finished writing’ (1) cannot be decomposed into two different verbs: (a) likhe ‘having written’, and (b) pheleche ‘he has thrown off’, because, as we can see, the basic meaning of the verb pheleche is non-existent in that Compound verb.

ii. Like words, Compound and Complex verbs have specific meaning. For example, the Conjunct verb bazar koreche in (3) can only mean ‘he has done shopping’, and cannot have a meaning like, for example, ‘he has built a market’. Equally a Conjunct verb like phon kOra means ‘to make a phone call’ and not ‘to fabricate a phone set’. Again, the conjunct verb bari kOra ‘house do’ means ‘to make/build a house; to be/become owner of a house’, whereas the verb phrase bari banano ‘house build’, means ‘to build/construct a house (preparing it for habitation)’.

iii. Unlike phrases (5), Compound (6) and Conjunct (7) verbs resist deletion of one of their constituents, which points to the fact that they have some sort of formal integrity similar to words.

iv. As we can see in Table 1, not all vector verbs can combine with all verbs or all verbals. ^[6] According to Dasgupta (1977: 70) the incompatibility between particular vector verbs and pole verbs is due to their (sub)categorial mismatch or ‘transitivity harmony’. Although Dasgupta’s rule of thumb works in a good number of cases, exceptions abound. For example, in the Compound verb shune jawa ‘continue to listen’ a transitive verb shuna ‘listen’ combines with an intransitive verb jawa ‘to go’. Equally, ghumiye neya ‘to have slept for a while’ is a combination of the intransitive verb ghumano ‘to sleep’ and the ditransitive verb neya ‘to take’.

v. As in other South Asian languages (cf. Masica 1976: 143) the meaning of complex predicates in Bengali is varied and manifold. Depending on the verbs involved and on the circumstances in which they are used, they can connote different “manners of action” (not to confuse with aspect) such as Completion (1), Forethought (8), Continuation (9), Benefaction (10), Directionality (11), Stubbornness (12), Regret (13), etc. (see Masica ibid. for a more exhaustive list of such manners).

vi. The meaning of Complex predicates may also depend on pragmatics. For example, the verb boshe achi ‘having.sat I.am’ may mean either ‘I am sitting’ or ‘I am waiting’ depending on the circumstances in which the verb is used.

As a whole, much idiosyncrasy and connotation is involved in complex predicates and syntax does not seem to be the right place to handle such things. Due to their idiosyncratic character and numerous connotations, it is difficult to consider Complex predicates as syntactic constructs (such as sentences and phrases), which are usually semantically transparent and compositional. If complex predicates cannot be handled satisfactorily in syntax, then one should try to account for them in morphology. If that also fails, then each one of these predicates has to be listed in the lexicon, like idioms, for example.

3 Complex predicates: Are they formed in morphology?

Models of morphology can be put into two different categories: Atomistic and Holistic. By ‘atomistic’ we refer to models that must involve units smaller than the word, such as affix or stem, whereas a true holistic model involves none of these units in the process of word-formation. In what follows, after a brief description of WWM (elaborated in Ford et al. 1997; Singh 2006), a holistic model of word-formation, we will try to demonstrate how complex predicates can be handled in the light of this model.

According to Singh (2006: 578):

Singh (2006: 578) states that

For WWM, words have no internal (non-phonological) hierarchical structure. According to this model (cf. Singh 2006: 578): “Morphological complexity is a matter of analyzability (# segmentability) of a word into a variable and a constant component with respect to a WFS.”

According to Singh (2006: 578) WWM sees “morphology” “not as a combinatorics of morphs or morphemes but as a system of generalized and abstract bidirectional correspondence among patterns instantiated by sets of whole words that exploit the same contrast.

For example, on the basis of morphologically related sets of words like (14) and (15), one can obviously set up a WFS like (16) which is licensed by a set of semantically related word-pairs that manifest the same (i) formal contrast: X/Xness on the one hand, and (ii) categorial affiliation: Adjective/Noun on the other. According to (16) kind, bold, bright, etc. provide the differing values for the variable X in (14), (15) while the phonemic sequence ness, which remains constant throughout (15), provides the particular value of the prime (′) in the schema (S1). ^[7]

Subcomponents can be represented by any phonic element: single phoneme, meaningless sound cluster, simple or complex word, discontinuous and inseparable segmental as well as supra-segmental means like stress and tone (variables, however, cannot be exclusively supra-segmental). For example, if the Hebrew word hagdala ‘enlargement’ is mapped onto (17), the variable will be represented by the discontinuous sequence of consonants: /h/-/gd/-/l/ and the constant by the discontinuous sequence of vowels: /a/-/a/-/a/.

Similarly, if ts^hu³⁵ ‘a plough’ is mapped onto (18), a WFS of Chinese, the constant will be represented by the rising tone and the variable with the sequence of segments. Quite coincidentally, when a subcomponent is represented by a continuous sequence of phonemes (ness for example) it can have phonic resemblance to word parts which some grammatical traditions would label as morpheme, root, stem or affix.

We shall now move onto demonstrate how WWM can account for complex predicates. We assume that the lexicon of a Bengali speaker contain sets of word-pairs constituted of verbals like likhe or bOle and also verbs like likhejay and bOlejay. Such pairs license (19) or (20) which can be activated, if needed, to form, analyze and retrieve other Compound verbs like shunejay ‘he continues to listen’ or dekhejay ‘he continues to see’ by mapping verbs like shune ‘having listened to’ or dekhe ‘having seen’ onto the WFS. Each pair in (19, 20) manifests semantic relatedness between the pair-mates (between likhe and likhejay for example), each pair-mate belongs to a different morphological category, and finally, in both of the pairs we can see the same formal contrast represented by the sequence jay. The Compound verb likhejay can be analyzed into the variable likh and the constant ejay if we map it onto (19). However, the same verb can be analyzed into the variable likhe and the constant jay if we map it onto (20).

Equally, the Conjunct verb bazarkore can be analyzed into the variable bazar and the constant kore respectively with (21). As we mentioned above, the subcomponents (variable and constant) are represented by a mere sequence of phonemes and the fact that likh is isophonic with the regular word likh ‘write’ or that the constant ejay is not isophonic with any regular word, is irrelevant for this model of morphology.

4 Bengali complex predicates: Are they words?

One may argue that Bengali complex predicates are not words but phrasal constructions because a large number of them (e.g. 22, 23) lack formal cohesion in the sense that their pole and the vector part can be interrupted with some other words, and hence they lack formal integrity, one of the characteristics of words.

Dasgupta (1977), however, claims that the constituting elements of Compound verbs cannot be separated from each other. His rule-of-thumb (1977: 75) is that if it is possible to insert an adverb between the pole and the vector, as in (24), then we are dealing with a sequence of poles (in other words, a serial verb), and if the particular construction shuns such insertions, then it is a Compound verb (25).

Although words are expected to have both (i) formal integrity and (ii) semantic opacity, some languages have been reported to have words that allow insertion of lexical and grammatical elements (cf. Sadock 1998). For example, in (26) the adjective bhalo can be separated from the rest of the subcomponents of the word bhalobashe ‘he loves’. Despite this separability, bhalobasha remains one single word because the sequence basha is completely devoid of meaning as a separate word in Modern Bengali. ^[8] Such examples point to the fact that words may lack formal integrity. On the other hand, linguists generally do not disagree with the Frege-Wittgensteinian idea that words lack semantic compositionality. Hence, semantic opacity seems to be a more reliable and universal criterion for word-hood as compared to formal integrity. In other words, words are expected to have formal integrity but semantic opacity is a requirement for word-hood (which does not mean that other linguistic constructs that manifest semantic opacity, for example idioms, are words). As complex predicates categorially lack semantic compositionality, they are words, and like any other word, they are better handled in morphology.

Although some Compound and Conjunct verbs happen to lack formal integrity, it seems that most of them must acquire it in the course of time. In some dialects of Bengali there are examples of Compound verbs (27, 28) that have been fused into simple verbs. (Conjunct verbs have not been noticed to have undergone such cohesion however.) However, a large number of Compound verbs are resistant to structural disruption (6, 7 and 25). It is possible that this resistance develops in order to prepare the way towards the formal cohesion that is finally achieved by some phonological change, and then by internal sandhi, at some point of development of the language.

Chittagonian

Dialect of Kishorganj

Only a small subset of regular verbs (we have listed about 20 of them in Table 1) appear as vector in complex predicates in Bengali and also cross-linguistically (see Butt 2003), which points to the fact that verbs that are used as vectors have a different distribution compared to other regular verbs that are used only as poles. Complex predicates are not different from the words known as compounds, except for the fact that the latter categorially manifest formal integrity. The vector part of complex predicates are like house in doghouse (29), which has a different distribution as compared to dog which can be replaced by any noun denoting animal whereas house cannot be replaced by most nouns denoting ‘place of living’ such as room, home, village, etc. Hence tiger-house, snake-house, etc. are acceptable, but *dog-room, *dog-home, etc. are not. In the same way, only some of the twenty vector verbs (phele ‘gives’, rakhe ‘keeps’, etc.) of Table 1 can replace the vector boshe ‘has sat’ in likheboshe (30) whereas many verbs can replace the pole likhe ‘having written’.

We claim, on the basis of (27, 28), that verbs that appear as vectors in complex predicates have already stepped into the process of grammaticalization and are heading towards becoming something that atomistic models generally describe as affixes – like dom in kingdom and hood in boyhood which were regular words at some point of diachrony. (This however does not mean that all vector verbs will cease to be used as regular verbs in the long run.) The process of grammaticalization involves, usually in consecutive stages, the loss of wordhood, loss of meaning, formal change and fusion with other subcomponents.

The Bengali verb/noun bhalobashe ‘he loves’ is a good example of what the process of grammaticalization may do to a word. In Middle Bengali bashe ‘he feels’ was used as a regular verb and also as a vector (see Chatterji 1970 [1926]: 899 for various uses of bash). In Modern Bengali, however, the regular verb use of basha is obsolete; it is devoid of any meaning (or “use” in the Wittgensteinian sense) and it appears exclusively in the verb/noun bhalobasha ‘to love/love’ and its paradigms. ^[9]

5 Problems with possible atomistic accounts of complex predicates

In this section, we will see whether atomistic models can satisfactorily handle all types of complex predicates. We have mentioned above that the vector part of the complex predicates may be considered as (becoming) suffixes or suffixoids that denotes manner (# aspect) of the action referred to. One of the options would be to list vector verbs and their regular verb counterparts as separate lexical entries. Vector verbs functioning as suffixoids are attached to the pole through some process after Aronoff (1976) or Lieber (1992). This is in keeping with Butt’s (2003) claim that vectors evolve into auxiliaries and then into clitics before becoming affixes.

Although this seems to work in a large number of cases, there are examples like bhalobasha ‘to love’, Ter pawa ‘to be aware of’, khabi khawa ‘to gasp (as a drowning person does)’ which are difficult to handle in these models. In bhalobasha the sequence bhalo is isophonic with the regular word bhalo ‘good’, but the rest of the sequence: basha (or bashi in bhalobashi ‘I love’) is neither a word nor a stem, because, as we mentioned above, it is devoid of meaning. ^[10] It cannot be a suffix either because no other word is formed with this submorphic unit. We note that, contrary to Butt’s (2003) claim, the vector in the verb bhalobasha as well as the vector in the complex predicates in (27, 28) have not evolved into some affix but was rather fused with the pole.

Equally, in Ter pawa or khabi khawa the sequences pawa and khawa are isophonic with regular verbs, but the sequence Ter or khabi are devoid of meaning. These constructions can be considered neither as an idiom nor as a phrase because both of these categories usually require each of their constituting elements to have a meaning. Be that as it may, things like basha, Ter or khabi should be problematic for atomistic models which must give a name to, and/or list each and every word-internal part.

In a WWM description these are simply sequence of phonemes, and the fact that we find them in the verb bhalobasha, Ter pawa or khabi khawa is a mere coincidence. These verbs can be output with WFSs like (31, 32), which, as we can see, may be licensed, among others, by pairs constituted of simple predicates. ^[11] We note that both simple and complex predicates can be formed with WFSs instantiating the schema /X/_a ↔ /X′/_b, and therefore, from a morphological point of view, the idea of the complexity of predicates turns out to be an exclusively (atomistic) theory-internal matter.

Although we have used traditional terms like pole, vector, complex predicate, Compound verb, Conjunct verb, etc. throughout this article to ease discussion, there will in fact be no need of using them if we adopt the WWM framework. Complex predicates are verbs, and like any other verb (or any other word), they can be formed, analyzed or retrieved with relevant WFSs, and therefore, neither they nor their subcomponents require different names.

6 Conclusions

In this article, we have claimed that constructions known as ‘complex predicates’ are in fact words. They can be satisfactorily handled within morphology, and therefore, there is no need to treat them in syntax, or list each one of them in the lexicon, or put them in some special category like Word+ after Kageyama (2001). The so-called vector verbs and their regular verb counterparts do not need to be listed separately either. However, depending on the speaker-hearer a number of these complex predicates must be part of his individual lexicon, which together with other words form adequate pairs that license WFSs like (19, 21) or (30, 31). Once such WFSs are part of his morphological module, the speaker-hearer can activate them, as needed, to output, understand or retrieve other complex predicates