Adjective Classes. Adjective lexical-semantic classes, like nominal classes, are less studied than verbal ones, although there is some recent work on adjective classes. Xxxxxxxxxxxxxxxx and XxXxxxx (1993) identified adjective scales, e.g. hot-warm-cold, by using cues for scalar adjectives and then clustering adjectives, and evaluated against human-created clusters. Xxxxxx et al. (2007) presented a clustering algorithm based on word sense induction to cluster adjectives in Japanese and English, evaluated against lexical resources such as WordNet. Navarretta (2000) clustered Danish adjectives based on predicative patterns, followed by manual editing of the clusters. Xxxxxx and Xxxxxxx (2003) performed unsupervised acquisition of Catalan adjective classes, with the resulting clusters evaluated by human judges. Boleda (2004) focused on unsupervised clustering of Catalan adjective semantic classes by exploiting a range of shallow distributional linguistc features. In D6.1 UCAM intended to work on noun classes. However, an opportunity arose to research classification of adjectives rather than nouns. As a result, an MPhil dissertation supervised by PANACEA members has been completed on unsupervised adjective clustering. This work uses spectral clustering to group adjectives using syntactic features, specifically subcategorization patterns, along with co-occurrence and semantic features. Evaluation is against a novel gold standard based on Xxxxx (1991) and the F-score of the best feature set is 58.
