Compositional Distributional Semantics with Syntactic Dependencies and Selectional Preferences

This article describes a compositional model based on syntactic dependencies which has been designed to build contextualized word vectors, by following linguistic principles related to the concept of selectional preferences. The compositional strategy proposed in the current work has been evaluated on a syntactically controlled and multilingual dataset, and compared with Transformer BERT-like models, such as Sentence BERT, the state-of-the-art in sentence similarity. For this purpose, we created two new test datasets for Portuguese and Spanish on the basis of that defined for the English language, containing expressions with noun-verb-noun transitive constructions. The results we have obtained show that the linguistic-based compositional approach turns out to be competitive with Transformer models.

SELECTIONAL PREFERENCES OF THE SUFFIX -IFY IN NAMES OF SOME MOBILE APPS IN ENGLISH

We are witnessing an increase in the number of formations with the verb-forming suffix -ify, the majority of which are names of websites, mobile apps or internet-based services, e.g. Androidify, Connectify or Fatify. The aim of this paper is to provide insight into the selectional behaviour of -ify in these new formations and determine whether there are any indications of some new trends. The suffix -ify typically occurs with Latinate bases to form causative or inchoative verbs such as intensify or mummify (Bauer, Huddleston 2002; Plag 1999). The rivalry with -ize is mainly resolved by phonological constraints: -ize occurs after one or more unstressed syllables, while -ify is always preceded by a stressed syllable. This means that -ify attaches to either monosyllabic bases or polysyllabic bases stressed on the ultimate (Plag 1999, 2003). It is uncommon for neologisms with -ify to show stress shift (Plag 2003: 93). While some of the new formations are clearly compliant with these rules and tendencies, some are not. For example, there are some formations with non-Latinate bases (Dollify), some that require a shift of stress (Androidify) and some that attach to verbal bases (Distractify). By combining corpus search results (Corpus of Contemporary American English (Davies 2008-), Corpus of Global Web-Based English (Davies 2013), iWeb Corpus (Davies 2018-)), Google Play Store data and other sources (in particular Johnson 2014a), we investigate a list of 442 new words with -ify in order to ascertain whether the suffix -ify is changing as regards its selectional preferences described above.

Collocation and Selectional Preferences: A Frame-based Approach

Most of the research conducted into collocation and semantic frames has dealt with these phenomena separately. The study of collocation has not figured prominently in the research agenda of frame semantics, and frame semantics has only sporadically been used as an analytical framework for collocation. This article is a contribution to narrowing the gap between the two fields. It does so by addressing key issues in the design of a frame-based approach to collocation, with a special focus on the relation between collocational patterns and semantic valency, and by providing arguments for the efficacy of the frame-semantic theoretical apparatus in explaining verb-adjective links that are not accounted for by the existing models of collocation. The methodology combines lexicographic resources as well as quantitative and qualitative analysis of examples and data from an English web corpus (ukWaC).

Syntactic composition and selectional preferences in Hindi Light Verb Constructions

Previous work on light verb constructions (e.g. chorii kar ‘theft do; steal’) in Hindi describes their syntactic formation via co-predication (Ahmed et al., 2012, Butt, 2014). This implies that both noun and light verb contribute their arguments, and these overlapping argument structures must be composed in the syntax. In this paper, we present a co-predication analysis using Tree-Adjoining Grammar, which models syntactic composition and semantic selectional preferences without transformations (deletion or argument identification). The analysis has two key components (i) an underspecified category for the nominal and (ii) combinatorial constraints on the noun and light verb to specify selectional preferences. The former has the advantage of syntactic composition without argument identification and the latter prevents over-generalization, while recognizing the semantic contribution of both predicates. This work additionally accounts for the agreement facts for the Hindi LVC.