Status : Doctorant
LLF, CNRS – UMR 7110
Université Paris Diderot-Paris 7
Case 7031 – 5, rue Thomas Mann,
75205 Paris cedex 13
Mail : foravnzvar[nebonfr]yvathvfg[cbvag]havi-cnevf-qvqrebg[cbvag]se
Website : http://www.llf.cnrs.fr/fr/Gens/Beniamine
Title : Inflectional classifications. A quantitative study of paradigm structures
PhD Defense : 2018-07-06
Inscription : 2014 à Paris 7
In some inflectional systems, the same morphosyntactic properties can be expressed differently across lexemes. These systems are usually described through the enumeration of a small number of inflection classes partitioning the inventory of lexemes. However, the actual structure of inflection class systems is much more complex, and methodological vagueness leads to contradictory accounts regarding inventories of inflection classes.
This dissertation adopts the Word and Paradigm approach and elaborates computational tools to investigate precisely the similarity structure of inflection class systems based on inflectional lexicon. We study Arabic, Yaitepec Chatino, Zenzontepec Chatino, English, French, Navajo and European Portuguese verbs as well as Russian nouns.
The first part defines the inflectional behavior of lexemes through the set of all surface alternations between their forms. We describe an algorithm to infer automatically alternation patterns between any two forms of a lexeme. We use alternation patterns to quantify the Paradigm Cell Filling Problem (PCFP). The second part investigates the similarity structure of inflectional systems. We start by classifying lexemes into microclasses, based on identity of inflectional behavior. These classes are numerous, and sometimes very similar. We then describe an algorithm based on minimal description length to gather microclasses into macroclasses which conform to the traditional notion of inflection class. Finally, we show that the most faithful model to describe similarities in inflectional systems is a lattice in which each node is an inflection class. To deduce this multiple inheritance hierarchy from alternation patterns, we use Formal Concept Analysis.