Implementational issues

In our HunGram implementational LFG grammar (Laczkó & Rákosi 2008-2013), we

“translate” É. Kiss’ (1992) GB analysis of Hungarian finite sentences into LFG-theoretic terms, and we implement this by employing a whole range of labelled constituent structure categories, as is customary in the practice of the ParGram enterprise. As I pointed out above in passing, É. Kiss’ (1992) model is unorthodox with respect to some crucial GB assumptions:

(i) it employs the exocentric sentence label S (ii) it does not use verb-related functional categories at all (not even IP) (iii) it postulates a verb-initial, flat, free-word order V’, also containing the subject argument, and, consequently, it violates both the strict binary branching principle and the structurally asymmetrical representation of the subject and the object, and, even more generally and more seriously: there is no designated configurational encoding of these grammatical functions even in deep structure, which is alien to all mainstream Chomskyan models. Needless to say, all these marked aspects of É. Kiss’ GB account are absolutely unmarked for an LFG approach. Thus, at the LFG-theoretic level, in Laczkó &

Rákosi (2008-2013) we adopt É. Kiss’ unorthodox GB analysis in LFG in a fundamentally orthodox fashion in the above respects. There are, however, three aspects of our approach, which are not in accordance with standard LFG c-structural representational principles.

88 Eventually, it may turn out that it is only verbs (VPs) that call for, or admit, this augmentation cross-linguistically.

89 On the basis of (117), subject and/or topic seems even more appropriate than subject/topic in (116).

90 On this scenario, the following three parametric options seem to emerge across languages: (i) strictly VP-external subject (English) (ii) VP-internal subject in a designated position (Russian) (iii) VP-internal subject without a designated position (Hungarian).

(A) If a language is analyzed as making use of the [_S NP XP] configuration then the default assumption is that the NP has the SUBJ grammatical function, see Section 2.2 above. By contrast, in our analysis this constituent (potentially bearing a whole range of argument and adjunct functions) has the topic discourse function.

(B) The other unusual aspect of our approach in an LFG setting is that the specifier position of our VP (at least optionally) hosts a focused constituent. This assumption is in contrast with the rather generally accepted view that it is the specifier positions of functional categories that typically host constituents with discourse functions, and the VP is a lexical projections, see the discussion of Bresnan (2001) and Dalrymple (2001) in Sections 2.2 and 2.4.2.2. For a treatment of Spec,VP as hosting focus, similar to our approach in this respect, see Mycock (2006).

(C) Another rather widely-accepted VP-related view is that VPs normally contain all the arguments of the verb except for the subject (also see point (A) above). By contrasts, we and Mycock (2006) assume that, unless it has a discourse function, the subject is also generated within the VP (in a flat, verb-initial V’).

When I am developing my LFG analysis of Hungarian finite clauses, I will defend the approach outlined above, with its general (and unmarked) and specific (and somewhat marked) aspects.

Let me also mention that the fact that É. Kiss (1994a) replaces the S symbol with T(ense)P makes that version one degree less adaptable in LFG, and the more functional projectional the MP analysis of Hungarian sentence structure becomes the further away it gets from adaptability in LFG. For the above reasons, we found É. Kiss’ (1992) model empirically, general (generative) theoretically, theoretic intuitively absolutely attractive (in our view it is a further advantage of É. Kiss (1992) and É. Kiss (1994a) over Brody (1990), (É. Kiss (2002) and many subsequent Chomskyan accounts that these two works capture the complementarity of VMs and focused constituents by assuming a competition between them for a singular designated preverbal position). In Laczkó & Rákosi (2008-2013) our aim was the HunGram (ParGram) implementation of the following LFG-style constituent structural representation of the most crucial aspects of a simple finite sentence in Hungarian. Compare (119) below with the skeletal representation of É. Kiss’ (1992) model in (9) in Section 2.1.1 repeated here as (118).

(118) CP

C E

XP S

XP* VP

[topic]

QP VP

Spec V’

[focus]

[VM] V XP*

(119) CP

C S

XP*[topic] VP

[ctopic]

[sent.adv.]

XP*[quant] VP

XP[focus] V’

PRT

(V ) X(P)*

The following remarks on (119) are in order here.

(c) É. Kiss (1992) assumes that when there is more than one quantifier constituent preverbally, they are individually and iteratively adjoined to the VP. As opposed to this, in her analysis ordinary topics and sentence adverbs are dominated by S in a flat structural configuration, and contrastive topics are treated as left-dislocated elements between C and S, dominated by the E(xpression) node. By contrast, in our implemented grammar not only quantifiers but also sentence adverbs and both types of topics follow the adjunction pattern, and the adjunctions of these three different categories can freely intermingle.

(d) As regards the treatment of the Spec,VP position, the current version of our grammar is rather limited. As is well-known, this position can be occupied by a great variety of categories of varying complexity, collectively (and loosely) called verbal modifiers (VMs) and, at least in several approaches (including É. Kiss (1992, 1994a) and ours), by focused constituents, and by wh-expressions (in complementary distribution); however, our grammar posits only a focused constituent or a preverb (a.k.a. coverb) belonging to VMs (no question expressions and no other types of VMs). We assume that the preverb (having the syntactic category PRT) is a nonprojecting word (in the sense of Toivonen (2001)).

From the complementarity of the two categories it also follows that a PRT can never be focused in our approach.

(e) The reason why the verb is also in parentheses in (119) is that our grammar also covers verbless clauses, containing NP or AP predicates. (In such cases there is no (empty) V positon in our c-structure representation, on the basis of standard LFG assumptions.

(f) The symbol X(P)* below V’ in (119) encodes that a nonprojecting word (a PRT, in particular) can also follow the verb.

(g) In a basic way, we also model VP- and V’-negation.

(h) Although we capture the generalized, LFG-style c-structural approach shown in (119), in our implementation of this grammar, following the practice of the implemented grammars of other languages in the ParGram community, we use a whole range of specific c-structure categories with “individuated”, mnemonic labels, which enhances the efficiency of the parser. The following figure gives a quasi hierarchical overview of the most important labelled categories. Notice, however, that they are employed to reduce the search space of the parser, and their ontological status is radically different from various functional projections in the Chomskyan paradigm.

ROOT

{ Sdisloc | Sfin }

{.|?|!|:|;|,}

Sdisloc Sfin

XPleftdisloc+

CPembed CPadj CPcond

CPrel

{ Sfinctopic | Sfin } (AdvConj) Sfinctopic Sfintopic Sfin_ADV VPquantneg

VPquant VPneg

VP

(C) Sfin XP (FOC)

PRT

Vbar Vbarneg (V) X(P)*

Figure 4. HunGram’s labelled categories Below are some remarks on this figure.

 The values of the ROOT rule has the following important function: if the parser is prompted to analyze a construction, and no specific phrasal category is given in the command then the parser will automatically attempt to analyze the string according to the categorial values specified in the rule. In addition, the rule also handles the punctuation marks at the end of the root categories.

 The two root categories in our grammar are: (i) complex sentences beginning with an embedded clause (Sdisloc, short for “dislocated sentence”) or ordinary finite sentences (Sfin), which themselves may contain embedded sentences.

 The Sdisloc version obligatorily contains at least one embedded sentence (CP) of various types (the + symbol stands for “at least one”, as opposed to the Kleene star, which means: any number, possibly null) and a finite clause either with a contrastive topic or without it: { Sfinctopic | Sfin }.

 A CP consists of an optional or obligatory C (depending on the CP type) and a finite clause (Sfin).

 A finite clause (Sfin) can be introduced by a conjunction (AdvConj) and it can have three major types: containing a contrastive topic (Sfinctopic), an ordinary topic (Sfintopic) or a sentence adverb (Sfin_ADV). (And, as I pointed out above, they can freely intermingle.)

 A finite clause may contain various types of VP projections with (different) adjunction properties.

 As I already pointed out above, a focused constituent and a PRT are in complementary distribution on Spec,VP.

 Sentence negation is possible even with a focused constituent. In this case the negative particle is between the focus and the verb. Our Vbarneg rule adjoins the negative particle to the V’ constituent.

 Above, I have also explained the parentheses around the V and around the P in X(P)*

below V’: in this way, we can handle verbless clauses, on the one hand, and the possible postverbal occurrence of a PRT, a nonprojecting word, on the other hand.

Some of the screenshots of the c-structures and f-structures of the sentences below exemplify the analyses of some basic finite clausal constructions in the current state of our grammar, Laczkó & Rákosi (2008-2013), while some others show that the treatment of certain (related) phenomena is incomplete or inappropriate.⁹¹

In Section 1.3 in Chapter 1 I showed our analysis of (58) repeated here as (120).

(120) Az okos fiú fel#hív-ott két szép lány-t.

the clever boy.NOM up#call-PAST.3SG.INDEF two pretty girl-ACC

‘The clever boy called up two pretty girls.’

I repeat its c-structure and its f-structure in Figures 5 and 6, respectively. For some comments on these representations, see Section 1.3.

Figure 5. C-structure of (120)

91 We have not carried out any joint development since 2013. The implementations I present in the rest of this dissertation are the results of my individual research: theoretical investigation and grammar implementation.

Figure 6. F-structure of (120)

Now consider the analysis of (121), in which the object is the focused constituent, and, therefore, the particle has to follow the verb.

(121) Az okos fiú két szép lány-t hív-ott fel.

the clever boy.NOM two pretty girl-ACC call-PAST.3SG.INDEF up

‘The clever boy called up TWO PRETTY GIRLS.’

Figure 7. C-structure of (121)

Figure 8. F-structure of (121)

As these representations demonstrate, the basic preverbal complementarity of particle VMs and focused constituents in the Spec,VP position is appropriately captured. However, consider the following example.

(122) János az asztal-ra tesz-i a toll-at.

John.NOM the table-onto put-PRES.3SG.DEF the pen-ACC

‘John puts the pen on the table.’

Figure 9. A correctly valid c-structure of (122)

Figure 10. A valid f-structure of (122), corresponding to Figure 9

Figure 11. An incorrectly valid c-structure of (122)

Figure 12. An incorrectly valid f-structure of (122), corresponding to Figure 9

(122) exemplifies a special type of VM: a maximally projected, referential oblique constituent. The Hungarian verb tesz ‘put’ is a three-place predicate in this use, and it requires its (goal) oblique argument to occupy the preverbal position (Spec,VP) in a neutral sentence, or else there must be a focused constituent in Spec,VP. The problem with HunGram’s treatment of (122) is twofold. On the one hand, it only provides the focused variant of the two appropriate analyses, see Figures 9 and 10, and there is no neutral sentence, VM analysis available. On the other hand, it erroneously produces an analysis as valid in which the Spec,VP position is not filled, see Figures 11 and 12. Here the oblique argument is analyzed as a topic, which is absolutely inappropriate. In Chapter 3, I will develop a comprehensive LFG-XLE account of the major types of VMs in Hungarian.

Next, consider the sentences in (123) and (124), and the analyses our HunGram produces in Figures 13-15.

(123) Ki hív-ja fel a fiú-t?

who.NOM call-PRES.3SG.DEF up the boy-ACC

‘Who calls up the boy?’

(124) *A fiú-t hív-ja fel ki?

the boy-ACC call-PRES.3SG.DEF up who.NOM

‘Who calls up the boy?’

Figure 13. The correct analysis of (123)

Figure 14. An incorrect analysis of (123)

Figure 15. Incorrectly valid c- and f-structures for (124)

Figure 13 shows that the current version of HunGram can analyze a ‘wh’-question appropriately: the ‘wh’-constituent must be focused, but Figures 14 and 15 show that it is not constrained enough to block analyses in which this constituent is not in Spec,VP (in Figure 14 it is analyzed as a topic and in Figure 15 it occurs postverbally). In Chapter 4, I will develop the LFG-theoretic and XLE-implementational aspects of the satisfactory treatment of ‘wh’-questions (including multiple ‘wh’ types).

Finally, consider the following examples containing predicate (i.e. clause) negation in (125) and (126) and our HunGram’s analyses in Figures 16-19.

(125) Kati nem hív-ja fel a fiú-t.

Kate.NOM not call-PRES.3SG.DEF up the boy-ACC

‘Kate doesn’t call up the boy.’

(126) *Kati a fiú-t hív-ja fel nem.

Kate.NOM the boy-ACC call-PRES.3SG.DEF up not ‘Kate doesn’t call up the boy.’

Figure 16. A correctly valid c-structure of (125)

Figure 17. A correctly valid f-structure of (125), corresponding to Figure 16

Figure 18. An incorrectly valid c-structure of (the ungrammatical) (126)

Figure 19. An incorrectly valid f-structure of (the ungrammatical) (126), corresponding to Figure 18

The analysis of (125) in Figures 16 and 17 is a correct one, and (in an appropriately modified form) I will keep it in my considerably augmented analysis of negation to be developed in Chapter 5. However, it is a shortcoming of the current state of HunGram that it does not yield another equally possible analysis of (125), in which Kati ‘Kate’ is not the topic but the focused constituent of the sentence. In Chapter 5, I will also make this alternative analysis available. In addition, as Figures 18 and 19 demonstrate, our current HunGram analyzes examples like (126) as grammatical sentences, in which the clause negating particle occurs after the verb, contrary to fact. In Chapter 5, I will rule this out.

2.5. Conclusion

Below, first I make general concluding remarks (Section 2.5.1), and then I add two implementational remarks (Section 2.5.2).

In document Laczkó Tibor Aspects of Hungarian Syntax from a Lexicalist Perspective (Pldal 119-133)