The semantics of meaning: distributional approaches for studying philosophical text

Distributional semantic models have risen to prominence in Natural Language Processing (NLP). Subsequently they have also been used in the digital humanities for studying conceptual change. This paper investigates the use of distributional models in philosophy. We propose a methodology for testing whether the models can be used for philosophical analysis. The methodology includes constructing a ground truth for philosophical terms, tuning distributional models for small data, learning embeddings for philosophical terms, and evaluating the learned embeddings using the constructed ground truth. We present results obtained with this methodology and show that Nonce2Vec, a model designed to operate on small text corpora, outperforms the more established Word2Vec in our evaluation framework. We also discuss some of the issues and potential pitfalls of applying distributional models to philosophical analysis