Analytic tenses in Slavic - an LFG approach

Andrew Spencer

Department of Language and Linguistics
University of Essex
spena@essex.ac.uk
http://privatewww.essex.ac.uk/~spena/


I first sketch the crucial aspects of the architecture of LFG. The
architecture of LFG distinguishes constituent-structure representations
from functional-structure representations. F-structure includes all and
only the interpretable morphosyntactic features, including grammatical
relations (SUBJ, OBJ etc.) and syntactic features which are interpreted
semantically such as [TENSE PAST], but not including purely formal
(morphological) features such as [VerbForm:Passive Participle]. The
c-structure is relative `surfacist' compared to some theories, though I
shall briefly mention more recent introduction of functional heads and the
notion of co-headed structures (from Kroeger, 1993, Bresnan 1996,
forthcoming).

I then present a survey of the Slavic past tense systems which have
developed from the old perfect forms realized by the auxiliary `be' and the
l-participle. I show how in most cases we have an analytic construction in
which the tense/aspect features in the syntax/semantics seem to be at
variance with their morphological realizations. A simple example is the
Czech simple past: napsal jsem dopis `I wrote(PF) a letter'.  The problem
here is that `Past Tense' seems to be realized by a form which elsewhere
(e.g. as a copula) would be `Present Tense' (similar cases have been
discussed in Romance within LFG by Schwarze 1996).

Börjars et al. (1997) have argued that LFG provides the best grammatical
architecture for handling analytic constructions such as these, and they
discuss the so-called `periphrastic' Latin perfective passive forms, in
which cells in the verb paradigm are filled by participle + auxiliary
phrases rather than individual word forms. I show that their approach has
to be modified by distinguishing crucially between m(orphological) features
such as [Tense:Pres] or [VerbForm:Pftv. Pass. Participle] and s(yntactic)
features such as [TENSE PAST] (as in Spencer and Sadler 1999). For this we
need a separate morphological component (m-structure) which manipulates the
m-features (I assume an "inferential-realizational" model of morphology,
after Stump's typology) and a special projection function which maps the
f-structure to m-structure (having roughly the effect of `Feature
Checking') (as in Butt et al. 1996, Frank and Zaenen 1998). In the Czech
example above the s-feature [TENSE PAST] is realized by the participle +
aux. construction as a whole. In the spirit of `Separationism' the
m-structure/f-structure relation does not have to preserve an isomorphism
between m-features and s-features, so the fact that the aux. is marked with
the m-specification [Tense Pres] is of no consequence. In standard accounts
based on unification which fail to distinguish m- and s- features (Börjars
et al., Minimalist Feature Checking etc.) the feature markings on aux. and
participle would fail to unify with the syntactic feature (or f-structure
AVM) and the derivation would crash.

The analysis raises a number of questions about the nature of syntactic
features, the nature of morphological features and morphological paradigms,
the morphology-syntax interface and the interpretability of
morphosyntactic features.