We perform a direct intrinsic evaluation of word embeddings trained on the works of a single philosopher. Six models are compared to human judgements elicited using two tasks: a synonym detection task and a coherence task. We apply a method that elicits judgements based on explicit knowledge from experts, as the linguistic intuition of non-expert participants might differ from that of the philosopher. We find that an in-domain SVD model has the best 1-nearest neighbours for target terms, while transfer learning-based Nonce2Vec performs better for low frequency target terms.

This research explores automated text classification using data from Low– and Middle–Income Countries (LMICs). In particular, we explore enhancing text representations with demographic information of speakers in a privacy-preserving manner. We introduce the Demographic-Rich Qualitative UPV-Interviews Dataset (DR-QI), a rich dataset of qualitative interviews from rural communities in India and Uganda. The interviews were conducted following the latest standards for respectful interactions with illiterate speakers (Hirmer et al., 2021a). The interviews were later sentence-annotated for Automated User-Perceived Value (UPV) Classification (Conforti et al., 2020), a schema that classifies values expressed by speakers, resulting in a dataset of 5,333 sentences. We perform the UPV classification task, which consists of predicting which values are expressed in a given sentence, on the new DR-QI dataset. We implement a classification model using DistilBERT (Sanh et al., 2019), which we extend with demographic information. In order to preserve the privacy of speakers, we investigate encoding demographic information using autoencoders. We find that adding demographic information improves performance, even if such information is encoded. In addition, we find that the performance per UPV is linked to the number of occurrences of that value in our data.

We evaluate the use of direct intrinsic word embedding evaluation tasks for specialized language. Our case study is philosophical text: human expert judgements on the relatedness of philosophical terms are elicited using a synonym detection task and a coherence task. Uniquely for our task, experts must rely on explicit knowledge and cannot use their linguistic intuition, which may differ from that of the philosopher. We find that inter-rater agreement rates are similar to those of more conventional semantic annotation tasks, suggesting that these tasks can be used to evaluate word embeddings of text types for which implicit knowledge may not suffice.