Semantics circle: "Small" vectors in big spaces, and the semantics of French -et(te)-suffixation

Code of Conduct for ZAS events: The ZAS is committed to fair, respectful, and professional interaction at its events. Therefore, please observe the Code of Conduct for this event.

Abstract

The talk investigates the behavior of the French "diminutive" suffix -et(te) through the lens of Distributional Semantics. The -ette variant of the suffix normally attaches to feminine bases to form feminine derivatives (see (1a)), and -et to masculine bases to form masculine derivatives (see (1b)). However, gender-mismatching derivations appear possible (see (2)), leading to a looser semantic relationship between base and derivative (Milner, 1989).

1. a. Boule (F)        Boulette  

     Ball            Small ball

b. Balcon (M)      Balconnet

     Balcony         Small balcony

1. a. Boule (F)         Boulet

     Ball               Cannonball

b. Char(M)                Charette

     Chariot                   Cart

Using word embedding models assigning words to dense, high dimensional vectors (Mikolov et al., 2013; Pennington et al., 2014 i.a.), we show that the distributional semantics of -et(te)-suffixation depends on the gender of the base, of the derivative, and crucially, on their interaction. Specifically, some (but not all) models capture the empirical observation that gender-matching base-derivative pairs in -et(te) are semantically closer than mismatching pairs, and that, within each group (matching/mismatching), pairs formed with the feminine suffix -ette tend to be semantically closer than those formed with its masculine counterpart -et. In terms of semantic stability, -et(te)-suffixation is shown to be more stable across gender-matching base-derivative pairs, than gender-mismatching ones.  Although their robustness should be further assessed, these findings are consistent with the hypothesis that -et(te) is ambiguous between a transparent, gender-matching diminutive modifier, and two heads with a looser, “similative” semantics, whose merger-site is gender-dependent.