Distributional and Multi-Modal Models for Compositionality andFigurative Language: German Noun Compounds and Particle Verbs

Sabine Schulte im Walde

Universität Stuttgart, Institut für Maschinelle Sprachverarbeitung

Distributional models assume that the contexts of a linguistic unit (such as a word, a multi-word expression, a phrase, a sentence, etc.) provide information about the meaning of the linguistic unit (Harris, 1954; Firth, 1957). They have been widely applied across data-intensive lexical semantics tasks (among other areas). In this talk, I will present ongoing work that explores distributional models and multi-modal extensions for automatically predicting compositionality and figurative language usage of German multi-word expressions, focusing on noun compounds and particle verbs. The first part of the talk will introduce standard distributional models to predict compositionality on the type level, comparing window- and syntax-based variants. The second part will present multi-model extensions by complementing textual with visual and associative information in count and predict vector representations. The third part takes a token-level perspective and presents a classification model combining textual and cognitive information for identifying figurative language usage of German particle verbs in context. Furthermore, I will zoom into empirical and lexical-semantic constituent properties from qualitative and quantitative perspectives, to investigate their influence on figurative language usage on the type and token levels.