Distributional Approaches to Semantic Relatedness: Generalisation, Evaluation, Visualisation

点击翻译按钮获取中文摘要

The project explores the potential and the limits of distributional approaches to lexical semantics within an interdisciplinary framework between theoretical, cognitive and computational linguistics. Relying on three test cases to distinguish various types of semantic relatedness, we bring together the tasks, approaches and results from the first project phase and study semantic relatedness from a meta-level perspective: We take "semantic relatedness" as a generic term subsuming paradigmatic relation pairs, compound-constituent pairs and sense relatedness across word types, and plan to (i) investigate distributional approaches across these types of semantic relatedness, across word classes, and across languages; (ii) explore the linguistic performance of soft clustering approaches, parameters, and evaluations to model ambiguity; (iii) exploit visualisation as a tool to assess semantic relatedness in distributional spaces; and (iv) abstract over semantic subcategorisation information and semantic evaluation for statistical machine translation as an extrinsic application. Across tasks and views, the distributional approaches continue to address two major challenges, (i) a theoretically and cognitively adequate selection of features to model word meaning and word relatedness, and (ii) modelling word meaning with respect to word senses, to address ambiguity.