Learning a Lexicalized Grammar for German

/ I Learning a Lexicalized Grammar for German Sandra Kfibler Computational Linguistics Gerhard-Mercator Universi~t Duisburg, Germany s. kuebl er@uni - duisburg, de Abstract In syntax, the trend nowadays is towards lexiealized grammar formalisms. It is now widely accepted that dividing words into wordclasses may serve as a labor-saving mechanism - but at the same time, it discards all detailed information on the idiosyncratic behavior of words. And that is exactly the type of information that may be necessary in order to parse a sentence. For learning approaches, however, lexicalized grammars represent a challenge for the very reason that they include so much detailed and specific information, which is difficult to learn. This paper will present an algorithm for learning a , link grammar of German. The problem of data spar- seness is tackled by using all the available infor- mation from partial parses as well as from an ex- isting grammar fragment and a tagger. This is a re- port about work in progress so there are no repre- sentative results available yet. 1. Introduction When looking at the most recent advances in syntax theory, one will notice a definite tendency towards lexicalized approaches. Simple context-free grammar formalisms may be easy to handle but they lack the descriptive power to model idiosyncrasies in the syn- tactic behavior of single words. In the natural language learning community, prob- abilistic approaches play a dominant role. Yet prob- abilistie learning has its strength in finding major trends in the training data. An idiosyncratic behavior of a single word is very likely to go unnoticed for lack of data. This divergence in interest might be the reason why hardly any attempt was made to have a lexicalized grammar learned. In this paper, I will describe an approach to learning a link grammar. Link grammar (Sleator & Temperley 199 I) is highly lexicalized, and therefore the problem of data sparseness will be immense. As a conse- quence, I have chosen a fuzzy representation. The fuzziness in this case models uncertainty rather than vagueness inherent in the language. The learning algorithm tries to extract as much information as pos- sible from a grammar fragment, partial parses pro- vided by this grammar, and wordclass information (for unknown words or to corroborate decision made by the system). 2. Link Grammar Link grammar (Grinberg, Lafferty & Sleator, 1995; Sleator & Temperley 1991) is a highly lexieal, con- text-free formalism that does not rely on constituent structure. Instead, it models connections between word pairs without building a hierarchical strueture. The link grammar formalism is best explained with an example of a linkage (i.e. a link grammar parse): Figure 1 shows a linkage for an English sentence. A linkage is a graph in which the vertices, representing the words, are connected by labeled arcs. These arcs are called links. For a grammatically correct sentence, the linkage must fulfill the following requirements: the links do not cross (= planarity), the graph is con- nected, and at most one arc connects a pair …

Content truncated to 500 words