This study compares two methods of topic detection, Latent Dirichlet Allocation (LDA) and Latent Semantic Analysis (LSA), by using it in conjuction with the Topic Context Model (TCM) on the task of keyword extraction. The surprisal values that TCM outputs based on LDA and LSA are compared, both, directly and as inputs to a Recurrent Neural Network (RNN). While in the direct comparison LSA slightly outperforms LDA, LDA and LSA perform on a par when a Recurrent Neural Network (RNN) is trained with surprisal values. In general: semantic surprisal as input of an RNN improves its performance.

This study focuses on the identification of English Idiomatic Expressions (IE) using an information theoretic model. In the focus are verb-noun constructions only. We notice significant differences in semantic surprisal and information density between IE-data and literals-data. Surprisingly, surprisal and information density in the IE-data and in a large reference data set do not differ significantly, while, in contrast, we observe significant differences between literals and a large reference data set.

Poster | BibTex | https://aclanthology.org/2023.konvens-main.15/