Automatic Acquisition of Artifact Nouns in French

Proceedings of Recent Advances in Natural Language Processing, pages 216–223, Hissar, Bulgaria, Sep 7–9 2015. Automatic Acquisition of Artifact Nouns in French Xiaoqin Hu Laboratory LDI University of Paris 13 franhuxqin@hotmail.fr Pierre-Andr´e Buvet Laboratory LDI University of Paris 13 pierreandre.buvet@gmail.com Abstract This article describes a method which al- lows acquiring artifact nouns in French automatically by extracting predicate- argument structures. Two strategies are presented: the supervised strategy and the semi-supervised strategy. In the super- vised method, the semantic classes of ar- tifact nouns are recognized by identifying the predicate-argument structures with the syntactic patterns of the given predicates. In the semi-supervised method, the ex- traction of predicate-argument structures is carried out from a semantic class of ar- tifact nouns given in advance. The pred- icate candidates obtained from extracted predicate-argument structures are then in- tersected. Next, the syntactic patterns of predicates are automatically learned by probabilistic calculation. With the ac- quired predicates and the learned syntac- tic patterns, more artifact nouns are identi- fied. 1 Introduction The difficulties for automatic acquisition of terms might come from the linguistic techniques, the computational techniques or the limits of the natu- ral language processing theory. Nowadays, many studies have been conducted for term extraction. This article presents a method for automatic ac- quisition of artifact nouns on the basis of syntactic- semantic analysis of predicates. Artifact nouns are the nouns of the artificial entities produced inten- tionally by human beings, with a view to a specific function. The automatic acquisition of artifact nouns is for completing the dictionary of semantic classes of laboratory LDI. There are two strategies for realizing this method: a supervised strategy in which the predicate-argument structures are ex- tracted by the syntactic patterns of the given predi- cates and a semi-supervised strategy developed on the basis of the supervised strategy. The semi- supervised strategy consists of two steps. In the first step, it predicts which predicates are relevant by a probabilistic calculation. In the second step, it appeals to the supervised strategy. This article is organized as follows. Section 2 states the re- lated work on term extraction in recent years. Sec- tion 3 presents the data model used in the proposed method. Section 4 explains in detail the proposed method including the semantic-syntactic analysis of appropriate predicates of artifact nouns. Section 5 presents the experiment results and the analysis of the results. 2 Related Work In the as-built systems of term extraction, the part of linguistic model is often limited to morpho- syntactic descriptions and the part of statistical model, to a large extent, depends on the statis- tical knowledge. TERMINO (Lauriston, 1994) and LEXTER (Bourigault, 1996), two well-known semi-automatic systems of term extraction, are based on syntactic descriptions. The method of Hearst (1992) and the method of Snow et al. (2004) take advantage of morpho-syntactic pat- terns for automatically recognizing hyponyms and hypernyms. The statistical methods for term ex- traction can be based on Markov model (Jiang, 2012), co-occurrence, or vector support, etc. ANA (Enguehard, 1993) is …

Content truncated to 500 words