Section: New Results

French FrameNet

Participant : Benoît Crabbé.

Elaborating on our previous work on Medieval French in collaboration with Sasha Simonenko (McGill) and Sophie Prévost (LATTICE), we have conducted the first large-scale quantitative investigation of the syncretisation of verbal subject agreement in this language and test a classic analysis which relates non-syncretic agreement and null subjects as parts of the same grammar. We have shown that agreement syncretisation and the emergence of overt pronominal subjects proceeded at the same rate. Under the Constant Rate Hypothesis of Kroch (1989), which states that a grammatical change has the same rate in different contexts, these results are compatible with the traditional analysis [40], [39], [33]. However, we show that this analysis also generates a number of predictions which are not borne out by the quantitative data. We conclude that a more complex model of interaction of subject and inflection parameters is needed. Such a model may for instance be one where the type of an ending (non-syncretic vs. syncretic), presumably dependent on some unrelated phonological mechanism, presents a parsing difficulty for a null subject-licensing grammar and thus lowers its probability to be chosen by the speaker, which eventually drives it to extinction, similarly to the grammar competition model proposed in Yang (2010).

We have also investigated the effects of the text form (prose vs. verse) on diachronic grammatical changes in Medieval French using parsed treebanks and (1 million words with PTB-like annotations). Despite the common intuition that the prose is somehow more “advanced” than the verse contemporary to it with respect to grammatical changes, the magnitude of the difference has remained unknown in the absence of quantificational evaluations. At the same time, the prevalence of verse in the earliest periods of documented French (i.e. X–XII c.) results in a strong and unavoidable correlation between time and form, which potentially undermines the results of the studies attempting to formally model Medieval French evolution. We have compared two historical changes across text forms (namely the loss of pro-drop and that of OVfinite order), and shown that verse and prose behave differently, at least regarding the OVfinite order, thus contradicting Kroch's (1989) Constant Rate Hypothesis [38].