On the Use of Automatic Tools for Large-scale Semantic Analyses of Causal Connectives

On the use of automatic tools for large scale semantic analyses of causal connectives Liesbeth Degand Université catholique de Lou- vain Institute of Linguistics degand@lige.ucl.ac.be Wilbert Spooren Vrije Universiteit, Amsterdam Language & Communication w.spooren@let.vu.nl Yves Bestgen Université catholique de Lou- vain Faculty of Psychology Yves.bestgen@psp.ucl. ac.be Abstract In this paper we describe the (annotation) tools underly- ing two automatic techniques to analyse the meaning and use of backward causal connectives in large Dutch newspaper corpora. With the help of these techniques, Latent Semantic Analysis and Thematic Text Analysis, the contexts of more than 14,000 connectives were stud- ied. We will focus here on the methods of analysis and on the fairly straightforward (annotation) tools needed to perform the semantic analyses, i.e. POS-tagging, lem- matisation and a thesaurus-like thematic dictionary. 1 Introduction In ongoing work, we explore the possibility to make use of large corpora to test hypotheses con- cerning linguistic factors determining the meaning and use of connectives. Of course, corpus-based approaches of connectives are not new, but classi- cally they consist of either fully analysed but rela- tively small corpora, or of large corpora of which a random set is analysed. The reason for this quanti- tative restriction is clear: The data-analyses are completely hand-based. While these empirical studies are useful from a qualitative point of view, they all suffer from the same quantitative draw- back, namely the relatively small number of data (rarely more than 100 occurrences are analysed, mostly only 50). In addition, most of these analy- ses are still too analyst-dependent, making gener- alizations and replications difficult. Changing this situation includes handling exhaustively large cor- pora (with hundreds and even thousands of occur- rences of the same linguistic phenomenon) and implementing the analytic procedures to make them analyst-independent. In this paper, we test such a methodology for which we used a number of linguistic hypotheses found in the literature on the semantics of causal connectives and tried to replicate the results. The linguistic material we worked on are four Dutch backward causal con- nectives: aangezien ('since'), doordat ('because of the fact that'), omdat ('because') and, want ('be- cause'). This choice was motivated by the fact that there has already been quite some linguistic work on this topic, mainly empirically based (Degand, 2001; Degand and Pander Maat, 2003; Pit, 2003).1 We have shown elsewhere how linguistic hypothe- ses concerning the scaling of these connectives in terms of subjectivity and their thematic behaviour could be supported (Bestgen et al., 2003). Since these first results are very encouraging, we would like to focus here on the methods of automatic analysis – Latent Semantic Analysis and Thematic Text Analysis - and on the fairly straightforward (annotation) tools needed to perform the semantic analyses, i.e. POS-tagging, lemmatisation and a thesaurus-like thematic dictionary. We illustrate how the combination of the two techniques of automatic analysis permit to gain deeper insight into the semantic constraints on the use of the con- nectives studied. Doing so, we test a number of new hypotheses concerning …

Content truncated to 500 words