Agreement in an XLE Grammar of Polish

Agnieszka Patejuk and Adam Przepiórkowski

Abstract

Introduction This abstract presents, on the basis of attested examples from the National Corpus of Polish (http://nkjp.pl/), a representative sample of agreement patterns modelled by the currently developed LFG grammar of Polish. It seems that some phenomena discussed below may pose a challenge to present approaches to agreement (such as the feature indeterminacy by underspecification account).

Data and implementation Let us start with attributive adjectives: Polish nouns are marked for case, number and gender and typically agree with their modifiers in these features:

(1) słucha świetnej muzyki i czyta dobre książki
listens great.gen.sg.f music.gen.sg.f and reads good.acc.pl.f book.acc.pl.f
'(He) listens to great music and reads good books.'

Some lexemes, however, display some interesting properties with regard to agreement:

(2) a. "Nowa Polska" uznana była za pismo ciekawe/*ciekawego
New.nom.sg.f Poland.nom.sg.f considered.nom.sg.f was.sg.f for mag.acc.sg.n interesting.acc/*gen.sg.n
'Nowa Polska was considered to be an interesting magazine.'
b. To zależy co kto uważa za coś ciekawego/*ciekawe
it depends what who considers for something.acc.sg.n interesting.gen/*acc.sg.n
'It depends on who considers what to be something interesting.'

In (2a), as in (1), adjectival modifiers fully agree with their head nouns in relevant features (Nowa with Polska, ciekawe with pismo). This is not the case in (2b) where coś and ciekawego bear different case marking, accusative and genitive, respectively. Sometimes, however, modifiers agree with coś:

(3) Wiesz o czymś ciekawym/*ciekawego?
know about something.loc.sg.n interesting.loc/*gen.sg.n
'Do you know of something interesting?'

On the basis of such examples, it seems that coś agrees with its modifier unless it is marked for nominative or accusative case - in such situations its modifier must be genitive. Furthermore, such patterns are not limited to coś: lexemes which display identical behaviour include co ('what'), cokolwiek ('whatever') and nic ('nothing'). These observations are formalised using the following agreement templates:

(4) a. agr-ng ≡ (↑ num)=(↓ num) (↑ gend)=(↓ gend)
b. agr-case-full ≡ (↑ case)=(↓ case)
c. agr-case-coś ≡ (↑ pred)∈c {co,coś,cokolwiek,nic}
{(↑ case) ∈c {nom,acc} (↓ case)=c gen | (↑ case) ∉ {nom,acc} agr-case-full}

Number and gender agreement between the head and its modifier are handled by (4a), while (4c) ensures case non-agreement with nominative or accusative heads and agreement elsewhere (via (4b)).

When discussing twists in Polish agreement, data provided by numerals must not be overlooked:

(5) a. Komu te kilka drzewek przeszkadza?
whom this.acc.pl.n several.acc.pl.n tree.gen.pl.n disturbs
'Who do these several trees disturb?'
b. Tych kilka drzewek jest bardzo ważnych.
this.gen.pl.n several.acc.pl.n tree.gen.pl.n is very important.gen.pl.n
'These several trees are very important.'

There are two types of Polish numerals: agreeing numerals, like typical nouns, trigger full agreement with their modifiers and finite verbs (when in subject position and marked for nominative). On the other hand, non-agreeing numerals such as kilka in (5), require default agreement (third person singular neuter) when in subject position. Furthermore, while they agree in number and gender with attributive adjectives, the latter may be marked for accusative (like the numeral, as in (5a)) or genitive case (like the noun, see (5b)).

In order to handle all agreement facts outlined above, the following templates are added to those provided in (4). (6a) uses the acm (accommodability) attribute which distinguishes numerals which agree in case with the accompanying noun (congr value) from non-agreeing ones (rec).

(6)a.agr-case-numrec ≡ (↑ acm)=c rec (↑ obj case)=(↓ case)
b.agr-case-normal ≡ (↑ pred) ∉ {co,coś, cokolwiek, nic}
{agr-case-full | agr-case-numrec}
c.agr ≡ agr-ng {agr-case-normal | agr-case-coś}

The template in (6c) handles all head-modifier agreement patterns described above: it ensures number and gender agreement by (4a), it calls (4c) with appropriate lexemes and (6b) elsewhere. The latter template handles case agreement with the head nominal via (4b) or allows the modifier to agree in case with the noun associated with the non-agreeing numeral (modelled as its object), which is formalised in (6a).

Let us now proceed to another interesting agreement phenomenon in Polish, namely agreement with predicative adjectives (for reasons of space, secondary predication cannot be discussed in this abstract):

(7) a. coś takiego jest potrzebne/*potrzebnego.
something.nom.sg.n such.gen.sg.n is needed.nom/*gen.sg.n
'Something like this is needed.'
b. tych kilka słów jest konieczne
this.gen.pl.n several.acc.pl.n word.gen.pl.n is necessary.acc.pl.n
'These several words are necessary.'

(2a) shows full agreement in number, gender and case between the predicative adjective, uznana, and the subject, "Nowa Polska". (7a) demonstrates that while attributive adjectives do not agree with coś when it is marked for nominative or accusative (as in (2b)), predicative adjective must fully agree with the subject. Non-agreeing numerals, on the other hand, trigger identical agreement patterns in these environments: both predicative and attributive adjectives may either agree with the numeral or with the accompanying genitive noun (its object). In (5b) both adjectives, predicative (ważnych) and attributive (tych), agree with the genitive noun (drzewek); in (7b) the predicative adjective (konieczne) agrees with the accusative numeral (kilka), while the attributive one (tych) agrees with the genitive noun (słów). Note that both adjectives, predicative and attributive, may choose the target with which they agree, the numeral or its object, independently of each other (as in (7b)).

The template provided below ensures appropriate agreement between the subject and the predicative adjective in examples discussed above (the "%s" variable hosts the agreement controller, "(↑ subj)" for instance):

(8) agr-pripred ≡ (%s num)=(↓ num) (%s gend)=(↓ gend)
{(%s case)=(↓ case) | (%s acm)=c rec (%s obj case)=(↓ case)}

The second disjunct in the second line of (8) allows the predicative adjective to agree with the object of an accusative non-agreeing numeral, while the first disjunct allows full agreement at all times.

Challenges to theoretical approaches The agreement phenomena presented above, while they are correctly modelled in the XLE implementation of Polish grammar, seem to provide certain challenges to current theoretical LFG approaches to agreement which choose to lexically specify the relevant features of the agreement controller.

The account of agreement which was adopted in Polish LFG assumes that adjectives bear their own morphosyntactic specification and accordingly agree or not in the relevant features with the agreement controller. This is formalised in the statements provided in (4), (6) and (8). While in the first two sets of statements, describing the agreement with attributive adjectives, the nominal head serves as the agreement controller, in (8) the predicative adjective agrees with the subject.

However, many (if not most) approaches to agreement seek to simplify the f-structure representation by placing appropriate agreement statements directly in the lexical entries. A typical example is the standard LFG treatment of subject-verb agreement where the lexical entry of the verb specifies the relevant features of the subject such as person, number and, if applicable, gender and case (see Dalrymple and Kaplan (2000, p. 763)). The same approach may be adopted for the purposes of agreement with attributive adjectives: modifiers do not bear their own agreement features, instead they directly specify the relevant features of the head (Dalrymple and Nikolaeva 2006, p. 839). A variation on this theme is described in Dalrymple et al. (2009): while it is considerably more powerful in using a complex case attribute and underspecification (in the lexical entries of nominals, in the specification of case of subcategorised arguments), it suffers from similar problems due to the way in which agreement is handled. Using inside-out equations (Dalrymple et al. 2009, p. 49), the lexical entries of modifiers specify negatively the case features of their head which are not realised by the modifier. In this way, modifiers may narrow down the underspecified morphosyntactic specification of nominals.

Non-agreeing numerals and coś-type items are problematic to approaches where agreement is ensured by imposing constraints on the values of appropriate features of the agreement controller in the lexical entries of modifiers. The unambiguously accusative numeral phrase in (5b) and (7b) is modified by a definitely genitive adjective, while in (2b) and (7a) the coś-type item, definitely nominative, bears a modifier which is genitive. On the account of agreement discussed above, the agreement controller and target impose conflicting case requirements, leading to ungrammaticality, contrary to fact. Finally, if this approach to agreement is adopted to predicative adjective agreement, similar problems are expected with non-agreeing numerals where the numeral phrase controller is accusative, while the predicative adjective target may be definitely genitive, as in (5b).

References