Challenging distributional models with a conceptual network of philosophical terms

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2511–2522 June 6–11, 2021. ©2021 Association for Computational Linguistics 2511 Challenging Distributional Models with a Conceptual Network of Philosophical Terms Yvette Oortwijn♣† Jelke Bloem♣ Francois Meyer∗ Wei Zhou♣ Antske Fokkens∗† ♣ Institute for Logic, Language and Computation, University of Amsterdam ∗Computational Linguistics & Text Mining Lab, Vrije Universiteit Amstersterdam †Department of Mathematics and Computer Science, Eindhoven University of Technology {y.oortwijn,j.bloem}@uva.nl, {pia.sommerauer,a.s.fokkens}@vu.nl francoisrmeyer@gmail.com, csunset14330@gmail.com Pia Sommerauer∗ Abstract Computational linguistic research on language change through distributional semantic (DS) models has inspired researchers from fields such as philosophy and literary studies, who use these methods for the exploration and com- parison of comparatively small datasets tradi- tionally analyzed by close reading. Research on methods for small data is still in early stages and it is not clear which methods achieve the best results. We investigate the possibilities and limitations of using distributional seman- tic models for analyzing philosophical data by means of a realistic use-case. We provide a ground truth for evaluation created by phi- losophy experts and a blueprint for using DS models in a sound methodological setup. We compare three methods for creating special- ized models from small datasets. Though the models do not perform well enough to directly support philosophers yet, we find that models designed for small data yield promising direc- tions for future work. 1 Introduction Philosophers apply text analysis to understand and delineate the precise meaning of concepts and the relations between them in a given text. This in- cludes comparative research that investigates dif- ferences in how concepts are viewed in differ- ent philosophical schools or by individual philoso- phers. Betti and van den Berg (2014) point out that comparative research on concepts should follow a conceptual model approach. This approach states that we should not look at shifts of individual con- cepts in isolation, but rather address changes of a conceptual model as a whole. In such a system, relations between concepts are made explicit and comparative studies should identify how such re- lations change. Previous studies have shown that distributional methods can be used to support philo- sophical research by retrieving passages relevant to concepts in an author’s work (e.g., the concept of grounding within the work of Bernard Bolzano, van Wierst et al., 2016; Ginammi et al., 2020), but can we also generate distributional semantic (DS) models that are precise enough to identify differ- ences in concepts? This paper takes a first stab at addressing this question. In particular, we address the challenges involved in dealing with highly technical domain- specific terms that are defined in small corpora. As such, our use case has properties difficult for DS modeling, but typical for disciplines work- ing with comparatively limited data. We com- pare domain-specific embeddings created using Word2Vec (Mikolov et al., 2013a,b) and a count- based SVD model (Levy et al., 2015) to those cre- ated by Nonce2Vec (Herbelot and Baroni, 2017), specifically designed …

Content truncated to 500 words