A Case Study of a Free Word Order

A Case Study of a Free Word Order Vladislav Kuboˇn and Mark´eta Lopatkov´a and Jiˇr´ı M´ırovsk´y Charles University in Prague Faculty of Mathematics and Physics Czech Republic {vk,lopatkova,mirovsky}@ufal.mff.cuni.cz Abstract The paper aims at the investigation of free word order. It concentrates on the rela- tionship between (formal) dependencies and word order. The investigation is performed by means of a semi-automatic application of a method of analysis by reduction to Czech syntactically annotated data. The paper also presents the analysis of intro- spectively created Czech sentences demon- strating complex phenomena which are not sufficiently represented in the corpus. The focus is on non-projective structures, esp. those connected with the position of clitics in Czech. The freedom of word order is ex- pressed by means of a number of necessary shifts in the process of analysis by reduc- tion. The paper shows that this measure provides a new view of the problem, it is orthogonal to measures reflecting the word order free- dom based on a number of non-projective constructions or clitics in a sentence. It also helps to identify language phenomena that generally pose a problem for dependency- based formalisms. 1 Introduction The phenomenon of word order freedom plays an important role in natural language processing. The less rigid the word order, the bigger challenge for all kinds of parsing algorithms it constitutes. In this paper we are going to study the rela- tionship between (formal) dependencies – defined through analysis by reduction, a stepwise sim- plification of a sentence preserving its correct- ness (Lopatkov´a et al., 2005)– and word order. We want to gain better insight into the problem by means of the application of a semi-automatic procedure to syntactically annotated data. This method can verify the concept of the analysis by reduction (introduced in Section 2) against real data and, at the same time, it can shed more light on the relationships between these two syntactic phenomena, dependency and word order. Our goal is twofold: First, we focus on typical, ‘core’ (projective) word order. We are going to quantify how many sentences can be completely processed by a sim- ple analysis by reduction (i.e., sentences are (cor- rectly) reduced until only the predicate is left in the sentence). In order to be able to perform this task for a large volume of data, namely syntacti- cally annotated data from the Prague Dependency Treebank (PDT) (Hajiˇc et al., 2006)), we have developed an automatic procedure (requiring, of course, a subsequent manual checking) on a rela- tively large subset of the PDT data. The results are presented in Section 3. Second, we present an analysis of more ‘periph- eral’ word order. When it is impossible to per- form simple reduction (without violating the cor- rectness constraint), we strengthen the analysis by reduction by the concept of ‘shifts’ – word order modifications which help to preserve the correct- ness constraint. Such data provide a very interest- ing material for the analysis of individual linguis- tic phenomena involved in complicated sentences (Section 4). …

Content truncated to 500 words