Exploring the value space of attributes: Unsupervised bidirectional clustering of adjectives in German

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2839–2848, Osaka, Japan, December 11-17 2016. Exploring the value space of attributes: Unsupervised bidirectional clustering of adjectives in German Wiebke Petersen Düsseldorf University, SFB 991 petersen@ phil.uni-duesseldorf.de Oliver Hellwig Düsseldorf University, SFB 991 ohellwig@ phil-fak.uni-duesseldorf.de Abstract The paper presents an iterative bidirectional clustering of adjectives and nouns based on a co- occurrence matrix. The clustering method combines a Vector Space Models (VSM) and the results of a Latent Dirichlet Allocation (LDA), whose results are merged in each iterative step. The aim is to derive a clustering of German adjectives that reflects latent semantic classes of adjectives, and that can be used to induce frame-based representations of nouns in a later step. We are able to show that the method induces meaningful groups of adjectives, and that it outperforms a baseline k-means algorithm. 1 Introduction The research presented in this paper is part of a larger project which aims at the semantic analysis of adjectival modification of nouns in German in a frame-based approach. Its approach is to model the conceptual mechanisms underlying adjectival modification by decomposing the meanings of adjectives (A) and nouns (N) in frames, i.e. in recursive attribute-value structures. The most common modification process is that the noun concept bears an attribute whose value is restricted by the adjective. The exam- ples in (1) show that some adjectives are strongly associated with an attribute like the color attributes or the attributes of size or age. Some of these adjectives do not only require a special attribute but also a special noun concept (like the color adjective ‘blond’ that only applies to ‘hair’). For other adjectives like städtisch ‘urban’ it is less clear the value of which attribute they specify. (1) a. schwarzer black Ball ball // Stift pen b. blondes blond Haar hair c. fröhlicher happy Junge boy // Abend evening d. städtische cityadj Schule school e. kindlicher childadj Organismus organism f. schneller fast Fahrer driver As adjectives often merely restrict the value of an attribute that is associated with the adjective, as ‘black’ modifying the attribute color in (1-a) such that COLOR(ball) = black, attribute-value struc- tures are well-suited for the task of modeling adjectival modification (cf. Pustejovsky, 1995). How- ever, flat attribute-value lists are not sufficient, because ‘black pen’ can denote a pen that is black, or a pen that writes in black, COLOR(writing of the pen) = black. Similarly in (1-c), it cannot be the This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http: //creativecommons.org/licenses/by/4.0/ 2839 same HAPPINESS-attribute applying to both ‘boy’ and ‘evening’ in the same way. Relational adjec- tives like ‘städtisch’ in (1-d) express a relation between the denotation of N and of the root of A, al- though the kind of relation is rather vague, a.o. OWNER(school) = city, OPERATOR(school) = city, LOCATION(school) = city. In cases like (1-e) where the noun is relational (cf. de Bruin and Scha, 1988; Löbner, 2011), …

Content truncated to 500 words