Nominal Clitics and Constructive Morphology in Hindi
Proceedings of the LFG99 Conference
The University of Manchester
Miriam Butt and Tracy Holloway King (Editors)
This paper examines the establishment of grammatical and discourse functions of NPs by morphological marking in Hindi. I argue that a constructive morphology analysis of case marking can be generalized in Hindi to both case and discourse clitics. Nordlinger (1998) proposes constructive functions for case markers, whereby a lexical specification such as replaces the traditional c-structure annotation . Crucially, under her analysis, case markers also carry clause-level information about grammatical functions. I show that extending this mechanism to discourse clitics in Hindi can account for their ability to identify the clause-level discourse function of their host NP. The constructive analysis also accommodates incorporated forms, multiple embedding, and domain restrictions. However, as with constructive case (Butt and King 1999, Lee To appear), the `flattening' effect of the constructive function on discourse markers is restricted to f-structural syntactic discourse functions and does not apply to semantic effects.
Case marking in Hindi has been studied in greater detail in the literature than discourse marking (Mahajan 1990; Mohanan 1994; Butt and King 1999).(1) Furthermore, similarities in the licensing of case and focus have been observed in several studies, but they have been generally oriented toward structural licensing (Horvath 1995; Butt and King 1996). Here I examine morphological marking that is independent of positional licensing.
Traditionally in LFG, grammatical functions have been identified by functional descriptions stating relations such as . Nordlinger (1998), building on work by Andrews (1996) and Simpson (1991) amongst others, proposes constructive functions for case markers, such as , in order to account for various properties of case morphology in nonconfigurational languages. In this paper, the extension of constructive morphology to discourse marking captures the similar ways in which case and discourse clitics contribute clausal information about their hosts independently of configurational positions. The main aims of this paper will be:
In section 1, I give an overview of the behaviour of certain discourse markers, providing evidence that they are similar to case markers in being clitic-like. Section 2 provides a more detailed discussion of the syntactic properties of these clitics. In section 3, I present an analysis for Hindi case clitics using constructive morphology, and then show how this analysis can be extended to discourse clitics. Finally in section 4, I point out that although the `flattening' constructive function proposed here identifies a syntactic function at f-structure, clitics ultimately require a distinct semantic mapping to establish fine-grained scope interpretations (Dalrymple 1993, Andrews and Manning 1999).
The main discourse markers used in Hindi are listed in (1):2
Hii marks exclusive focus, similar in some ways to only in English. It identifies a particular member out of a possible set. Bhii indicates inclusive focus, including a particular element in an existing set.(3) Finally, to performs a topic contrasting function. More semantically and syntactically restricted clitics include tak and bhar.
For purposes of consistency and due to space limitations, I will restrict most of the discussion in this paper to the focus marker hii. The first three clitics listed above appear very frequently in conversational speech and perform a range of pragmatically and semantically restrictive functions. The definitions in (1) are somewhat coarse approximations of the complex functional range of each marker. In example (2) I show instances of their use.
In (2a) `Radha' is exclusively focused; in (2b) `Radha' is inclusively focused; and in (2c) the topic `candle' is contrasted with the new information in the sentence. These examples show that discourse clitics identify the constituent which they immediately follow as marked for certain discourse roles. The discussion here is restricted to the nominal domain, but it must be noted that discourse clitics may also modify non-nominal elements in a clause, including verbal elements and adjuncts. Although the full range of these uses cannot be adequately addressed here, these facts are in keeping with the general property of discourse markers as identifying the focus of a clause, whether nominal or not. In fact, the analysis provided here for the nominal domain could ultimately be generalized to include further uses of discourse markers, as I discuss briefly in section 3.5.
I follow the analysis of case markers in Mohanan(1994:60) and Butt and King (1999) in analysing discourse markers also as syntactic clitics rather than morphological affixes. Various distinctive characteristics provide support for such an analysis. First, case markers may take phrasal scope over conjoined nominals. Example (3), from Butt and King (1999), shows this difference between clitic and affix behaviour.
In (3a), the oblique affix cannot take scope over the conjoined nominal stems. In (3b), however, the single case marker can mark the conjoined stems. Discourse markers pattern like case markers in this regard. They can also take phrasal scope over conjoined elements, as shown in (4).
Secondly, pauses may intervene between nominals and their case markers (Mohanan 1994:60). Discourse markers show the same property. By contrast, it is impossible to insert a pause between a nominal stem and an affix such as -e in (3b). In addition to this, nominal agreement affixes also affect the stress pattern of a noun, whereas case clitics do not (Butt and King 1999:5). Again, discourse markers pattern like case markers in this regard and do not affect stress. Finally, discourse clitics and case clitics can be mutually reordered (although this is subject to dialectal variation). However, neither type of clitic can intervene between the nominal stem and regular affixes.
Assuming that discourse markers are also clitics, it is fairly straightforward to argue that, like case, they are of the type which attach to a constituent rather than ones which occupy a clausal position, for example, directly under S. Several phenomena may be taken as evidence for this assumption.
Firstly, discourse clitics may appear within NPs with correspondingly restrictive scope. If these clitics appear in a position directly dominated by S or IP, their NP-internal positioning and scope cannot be easily accounted for.
Secondly, discourse markers only take scope over constituents to their left. This is distinct from non-constituent discourse particles in other languages. Koenig (1991) distinguishes between adverb-like and clitic-like behaviour of focus particles cross-linguistically. In the English examples in (5), the particle `only' shows adverb-like properties as its position and scope are relatively flexible.
The focus marker hii in Hindi contrasts with this in exhibiting a stricter, constituent-like behaviour.
In (6a) and (6b) it is not possible for hii to take focal scope over any constituent that follows it. It can only focus the immediately preceding constituent, Maya=ne, as shown in (6c).
In order to consider the full range of positions that case and discourse clitics may occupy in NPs, an exception to the clitic generalizations listed in section 1.1 must first be taken into account. This is a set of forms in which hii shows signs of being incorporated into its nominal host.
Since hii follows the element it modifies, it commonly follows case markers too (Verma 1971). However, when hii does occur between pronominals and their case markers, it usually shows signs of incorporation (Koul 1990; McGregor 1995). I list the personal and demonstrative pronominal forms with incorporated focus in (7).
We can take hii to be incorporated in these forms based on several characteristics:
The focus morpheme interacts with morphological stress; various types of phonological reduction take place; nasalization of the /i/ vowel occurs in these forms but never in cliticization; and gaps occur in the paradigm of forms which show these characteristics.
To summarize the data discussed so far, I have argued that:
In this section I turn to various syntactic properties of both types of nominal clitics, including position with regard to phrasal boundaries, clausal cooccurrence, and domain restrictions.
As mentioned already, discourse clitics may attach to the right edge of the focused or topicalized constituent. The appearance of discourse clitics on an NP does not require any changes in word order, as seen in (10).
Unlike agreement affixes on nouns, discourse clitics do not contribute information to the f-structure of the NP but rather to the f-structure of the outer clause which contains the NP. They identify their NP as the TOP or FOC of the main clause. In this capacity, they resemble case clitics since they perform a clause-level function.
Case clitics must be adjoined to the right of the nominal head; this can be seen in (11).
(11b) and (11c) show that case cannot appear on a modifier inside the noun phrase, and (11d) shows that case cannot be multiply iterated in a single noun phrase. Discourse clitics, on the other hand, may attach to a wider range of constituents in the NP, as shown in (12).(4)
(12b) and (12c) contrast with (11b) and (11c) in allowing the focus marker to appear on modifiers. Note that (12d) involves focus marking which is morphologically incorporated into the demonstrative. Given a constructive analysis which I will describe shortly, the morphological rather than syntactic appearance of hii here does not affect the establishment of clausal focus.
Finally, although a wider range of positions is possible for the discourse markers, they cannot be instantiated more than once in the NP, as seen in (13).
This restriction on multiple discourse marking of a constituent resembles the case restriction in (11d). I assume for now that the coccurrence restriction derives from semantic incompatibility, rather than a structural restriction.
Even when focus occurs in an NP-internal position, as in (12b-d), the whole NP is identified as clausal focus in terms of syntactic behaviour. Certain word order and syntactic focus phenomena support this generalization.
One indication can be found in cases of multiple focii. There is a restriction on having two morphologically focused arguments in the same clause in Hindi unless the speaker resorts to very marked intonation in unusual contexts.(5) Given this restriction, a cooccurrence clash implies that hii is identifying two different values for a single focus function in a clause.
If a marker is within an NP, a regular (outside-in) function such as would not rule out such a cooccurrence. For example, if a modifier inside an NP has a clitic with the annotation , focus would just map as an NP-internal attribute, as the simplified f-structure in (14) shows. Consequently, if two focus markers appear within two NPs, each NP would simply have a focus attribute inside it and no clause level clash would be registered.
However, (15) shows that NP-internal discourse marking percolates up to the clausal f-structure. In these examples, the discourse markers occur on specifiers within NPs to show that in spite of being embedded inside NPs they are `visible' at the clause level and can clash.
In (15a), hii marks the possessive modifier of the subject and in (15b) it marks the possessive modifier of the locative. In (15c), despite the fact that the two discourse clitics are structurally within NPs, a clausal cooccurrence is registered, resulting in an infelicitous sentence.
The general classification of hii and bhii as types of focus and to as a type of topic predicts further that the former two cannot occur multiply but they may cooccur with to.
(16) supports this prediction, showing that a topic function and a focus function can legitimately cooccur. This contrast is a further indication that the clausal f-structure is sensitive to NP-internal discourse markers.
The syntactic status of focused NPs can also be verified by examining the interaction of clitic marking with the grammaticalized focus position in Hindi: the preverbal position (Butt and King 1996). When a part of an NP is focused, the entire NP may occupy this focus position.(6)
In (17a), the NP containing the focus marker is in situ. (17b) shows that the marked element alone cannot appear preverbally, but in (17c) the entire host NP can optionally occupy this position.
Finally, it is important to establish the syntactic domain of focus in Hindi. In other words, what is the limit beyond which cooccurrences are permissible? Based on the contrast in (18), I take the finite clause to be the domain within which restrictions on multiple focii must hold.
(18a) is infelicitous due to the occurrence of two identical focus values. This contrasts with (18b), which contains an embedded finite clause with focus.
A regular annotation for identifying the grammatical function of an NP in a c-structure, for instance , defines a path from the clausal f-structure down to the value of its SUBJ attribute.
In Nordlinger's use of constructive morphology for case, however, case-markers themselves constructively identify the grammatical relations of arguments to the verb. When a case clitic bears a constructive specification, it contributes information about the higher f-structure within which it is contained, via an inside-out (IO) function application.
For example, if a case marker bears the functional description , the whole expression represents an attribute-value pair which exists in the higher f-structure. The up arrow indicates that the nominal itself is the value of a attribute in a higher f-structure. The annotation defines a path outward, from the lexical item to the clausal f-structure.
Some of the features of Hindi grammar are in keeping with Nordlinger's original arguments in favour of a constructive approach to case. As a discourse configurational language, Hindi allows considerable freedom in argument positioning. It shows little evidence for configurationally licensed grammatical functions.
Furthermore, mood, aspect, and semantic information can be contributed by the presence of certain case markers. For example, the dative marker ko can imply specificity and the ergative marker ne has been argued to indicate control on the part of the subject.
In (19), the only distinction in interpretation between the two sentences is in whether the action was deliberate on the part of the subject. Annotations on case clitics have therefore been argued to contribute both syntactic and semantic information to the clause level (Mohanan and Mohanan 1994; Butt and King 1999; Lee to appear). This assumption is in keeping with Nordlinger's (1998:74) discussion of encoding semantic restrictions within the case marker.
Most importantly for the discussion here, a constructive view allows us to unify under a single analysis the shared patterns of "bottom up" function identification found in Hindi clitic behaviour.
The representation in (20) shows how case and discourse clitics on nominals can employ similar constructive functions to indicate the clausal function(s) of their NP hosts.
Case clitics identify the grammatical function of the NP, while discourse clitics identify which of a set of possible discourse functions the NP is associated with.
I have already indicated some important differences between the two types of clitics. Although they share similar lexical properties, they have distinct syntactic positions.
As (11) showed, case markers must cliticize only to the right edge of the NP. I follow Butt and King (1999) in assuming the structure for case given below, in which the case clitic serves as the head of the functional projection KP. On the analysis here, DP is the sister of K and specifier functions are within DP.(7)
Discourse markers, on the other hand, may adjoin to any part of the NP and in fact are not even restricted to nominal elements. To cover this range, I assume the simple structure in (22) for now:
The important difference here is that case clitics head their own functional projection while discourse clitics merely adjoin under their sister's category.
The set of examples in (23) correspond to the focus-bearing sentences I introduced in (12). These examples show that in spite of the repositioning of hii within an NP, the f-structure of the NP in each case is `flattened' in an identical manner due to the constructive function mapping of FOC.
In (23a), the focus marker is on the right edge of the whole NP; in (23b) it is adjoined to the numeral and in (23c) it is incorporated into the demonstrative pronoun. (23c) shows that the incorportated focus forms discussed in section 1.3 are equally accounted for by this analysis.
In each case, the NP is established as clausal focus in the f-structure because identifies the entire NP f-structure as the value of the outer f-structure's focus function. Consequently, all three c-structures share the single f-structure above.
The examples in (23) describe structures in which functional projections carry a focus marking. For these situations, the mapping of the constructive function is straightforward, as a direct path of allows access to the outer clausal f-structure. Multiply embedded constituents do not permit a direct mapping of the constructive function to the clause level.
It is important to first note that several speakers considered focus-marking on adjectives much less acceptable than focus-marking on determiners like possessives.(8) In the absence of a more comprehensive corpus study, the solution given here is somewhat tenative and subject to dialectal restrictions. However, it is an intuitive extension and can in fact accommodate the observed speaker variation.
Assuming that these cases are possible for some speakers, an intervening node annotated prevents a direct identification of focus at the clause level. Instead, the ADJ will be identified as focus within the NP, as shown in the `incorrect' f-structure in (24).
This example does not identify the NP as the focus of the clause because the intervening ADJ f-structure maps focus to the outer NP f-structure only.
However, (25) indicates that adjectives are no different than other NP sites in terms of clausal focus identification.
For speakers who allow (25a), (25b) is bad due to the identification of two focus values. This is the same effect as in (15c) earlier, suggesting that focus on adjectives must be equally visible at the clause level.
In addition, certain speakers' judgements indicate that focus in embedded infinitives also percolates to the finite clause level, disallowing other instantiations of focus marking in the entire clause. This phenomenon is shown in (26), where the embedded instance of hii clashes with a second use of the morpheme in the main clause.
Both of these embedded contexts - adjectival modifiers and infinitival complements - are problematic for a annotation because a second intervening node is annotated This node blocks percolation of the focus to the top f-structure of the finite clause itself.
I account for these provisionally with a simple extension of the analysis given so far. The annotation can be substituted with the inside-out functional uncertainty equation in (27) (Kaplan and Zaenen 1995; Dalrymple 1993; Bresnan 1999).
This description states that the end value of an unspecified string of attributes is also associated with a particular discourse function. Because of the uncertainty of the string, this can accomodate both NP-internal ADJ functions as well as embedded infinitives, in addition to the simpler cases.
Furthermore, the speaker variation in allowing or disallowing these multiply embedded focii can be accommodated with a simple distinction in the lexical entries of discourse clitics:
(28a) requires one or more attributes in the GF string. (28b) is equivalent to the original proposal of .(9)>/sup> An added advantage of this representation is that it can be extended beyond the nominal domain. As mentioned in the presentation of the initial data, discourse markers can also appear on non-nominal elements in a clause, particularly verbal morphemes. In such cases, the GF in the annotations suggested in (28) may itself be optional, allowing the annotation to reduce to a simple statement of .>sup>(10)
In order to ensure that the outer limit of the focus domain is the finite clause, I include a second functional description in the lexical entries of all discourse clitics stating the existential constraint in (29):
This is in keeping with Nordlinger's 1998:122) use of clause-level information specification. Synthesizing the latest two additions, the functional descriptions in (28) associate a DF with a GF, and the addition of (29) in the lexical entry states that the association must be such that the focus attribute is in an f-structure which also bears a tense attribute. This takes care of attribute strings which would otherwise stop short of the clausal f-structure in their mapping.
A prediction of this domain restriction is that one should not find stranded instances of discourse marked NPs. The example in (30) shows that this seems to be the case:
In response to the question "Who feeds the cat?" in (30a), just saying Raam is perfectly acceptable. However, if the marker hii is adjoined to Raam, then Raam=hii cannot be a complete utterance; it requires a tensed verb as in (b(iii)).
To summarize, this section has shown how discourse clitics may mark various parts of an NP and still identify the whole NP as the focus of the clause. Discourse clitics exploit constructive functions in a manner similar to case, and in some dialects appear to be restricted to one level of embedding. In those dialects that allow further embedding within the NP, functional uncertainty and the requirement that focus percolate to a tense-bearing f-structure ensures the appropriate mapping.
The discussion so far has specifically addressed syntactic functions. However, if we return to the examples in (12), repeated below in (31), we can see from the English translations that semantic scope differences actually emerge based on repositioning the clitic within the NP. Such NP-internal scope distinctions based on the position of discourse markers can be observed in many languages and are discussed specifically with regard to Hindi in Verma (1971:85) as well.
The constructive mechanism does not account for these scope differences. Precisely because of its radically flattening effect, the constructive mapping overgeneralizes focus and does not represent these scope and meaning differences directly at the f-structure.
Note that this type of distinction can be seen in various other cross-linguistic phenomena. One possible example is wh-feature percolation:
In (32), the wh-expression is contained within the constituent appearing in the clause-initial position McCawley (1988:477). There is a distinction between the semantic interpretation of the wh-subconstituent and the syntactic behaviour of the entire, containing constituent.
Returning to Hindi, we can observe rather subtle semantic distinctions in the reordering of focus and instrumental case clitics. Noguchi and Harada (1990) discuss a similar phenomenon in Japanese, in which the reordering of dake ("only") and de ("by") results in distinct semantic interpretations. They describe these as absolute (de-dake) and minimal (dake-de) restriction readings. I adopt this basic terminology for the examples below.
From the meaning distinctions in the English translations in (33), we can see that (33a) implies an absolute necessity restriction while (33b) means that a bicycle is minimally sufficient (but not absolutely necessary).
Reading these fine semantic relations directly off the f-structure is inadequate due to the `flattening' of the NP. Andrews and Manning (1999:11) discuss how the flattening of f-structure is necessary for certain syntactic associations, but is often an insufficient guide for semantic interpretation. Their examples include `concentrically scoped' modifiers and complex predicates, and they argue against the mediation of semantics by the f-structure. In their `subset' view, where projections represent groupings of information, certain attributes may be shared while others are restricted. The semantic distinctions arising in the data presented here calls for a similar treatment.(11)
I do not provide an account of the semantics of these clitics here. However, I suggest that rather than unifying all of the information into a single level, whether f- or s-structure, a comprehensive analysis must distinguish the syntactic mapping proposed in this paper from the fine-grained semantics alluded to in this section.(12) In other words, in the syntax the entire NP is constructively identified as clausal focus, regardless of position of discourse clitics within NP; however, semantically, meaning differences emerge based on clitic adjunction.
This paper has presented a preliminary attempt to account for the status of discourse clitics in Hindi as being parallel in many ways to that of case clitics. While further research is still necessary for a more complete account of the intricacies of clitic behaviour, I have argued in favour of the following generalizations:
The extension of constructive annotations to discourse clitics has two formal consequences that go beyond those of case.
McGregor (1995) cites a similar restriction on the positioning of `emphatic particles', however he specifically observes that this restriction is not necessarily adhered to strictly by all speakers.