Learning within- and between-word variation in probabilistic OT grammars

Aleksei Nazarov


This paper proposes a novel method of inferring diacritics for representing between-word variation (exceptionality) in Optimality Theoretic (OT) grammars (e.g., Pater 2000, 2010) that makes it possible to infer such diacritics in the face of within-word variation. Existing methods of inferring diacritics in OT (Pater 2010, Becker 2009, Coetzee 2009) are based in categorical grammar learning (Tesar 1995), which makes them unable to handle within-word variation. Existing methods of inferring probabilistic OT grammars (e.g., Boersma 1998) handle within-word variation well, but have no provision to distinguish exceptional from non-exceptional words, and are incompatible with the main idea in Pater (2010). I show that this latter idea can be made compatible with probabilistic grammars based on a case study from Hebrew (Temkin-Martínez 2010), so that both within- and between-word variation can be learned.


Exceptionality; Variation; Learning; Indexed constraints

Full Text:


DOI: http://dx.doi.org/10.3765/amp.v5i0.4253

Copyright (c) 2018 Aleksei Nazarov