Acquistion of the Morphological Structure of the Lexicon Based on Lexical Similarity and Formal Analogy

The paper presents a computational model aiming at making the morphological structure of the lexicon emerge from the formal and semantic regularities of the words it contains. The model is purely lexemebased. The proposed morphological structure consists of (1) binary relations that connect each headword with words that are morphologically related, and especially with the members of its morphological family and its derivational series, and of (2) the analogies that hold between the words. The model has been tested on the lexicon of French using the TLFi machine readable dictionary. 1 Lexeme-based morphology Morphology is traditionally considered to be the field of linguistics that studies the structure of words. In this conception, words are made of morphemes which combine according to rules of inflexion, derivation and composition. If the morpheme-based theoretical framework is both elegant and easy to implement, it suffers many drawbacks pointed out by several authors (Anderson, 1992; Aronoff, 1994). The alternative theoretical models that have been proposed falls within lexeme-based or word-based morphology in which the minimal units are words instead of morphemes. Words then do not have any structure at all and morphology becomes a level of organization of the lexicon based on the sharing of semantic and formal properties. c © 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. The morpheme-based / lexeme-based distinction shows up on the computational level. In the morpheme-based conception, the morphological analysis of a word aims at segmenting it into a sequence of morphemes (Déjean, 1998; Goldsmith, 2001; Creutz and Lagus, 2002; Bernhard, 2006). In a lexeme-based approach, it is to discover the relations between the word and the other lexical items. These relations serve to identify the morphological family of the word, its derivational series, and the analogies in which it is involved. For instance, the analysis of the French word dérivation may be considered as satisfactory if it connects dérivation with enough members of its family (dériver ‘derivate’, dérivationnel ‘derivational’, dérivable, dérive ‘drift’, dériveur ‘sailing dinghy’, etc.) and of its derivational series (formation ‘education’, séduction, variation, émission, etc.). Each of these relations is integrated into a large collection of analogies that characterizes it semantically and formally. For instance, the relation between dérivation and dérivable is part of a series of analogies which includes dérivation:dérivable::variation:variable, dérivation:dérivable::modification:modifiable, etc. Similarly, dérivation and variation participates in a series of analogies such as dérivation:variation::dériver:varier, dérivation:variation::dérivationnel:variationnel, dérivation:variation::dérivable:variable. 2 Computational modeling The paper describes a computational model aiming at making the morphological derivational structure of the lexicon emerge from the semantic and the formal regularities of the words it contains. A first experiment is currently underway on the lexicon of French using the TLFi machine readable dictio-