Downsampling Strategies are Crucial for Word Embedding Reliability

The reliability of word embeddings algorithms, i.e., their ability to provide consistent computational judgments of word similarity when trained repeatedly on the same data set, has recently raised concerns. We compared the effect of probabilistic and weighting as downsampling strategies. We found the latter to provide superior reliability while being competitive in accuracy when applied to singular value decomposition-based embeddings

JeSemE: A Website for Exploring Diachronic Changes in Word Meaning and Emotion

We here introduce a substantially extended version of JeSemE, a website for visually exploring computationally derived time-variant information on word meaning and lexical emotion assembled from five large diachronic text corpora. JeSemE is intended as an interactive tool for scholars in the (digital) humanities who are mostly limited to consulting manually compiled dictionaries for such information, if available at all. JeSemE uniquely combines state-of-the-art distributional semantics with a nuanced model of human emotions, two information streams we deem beneficial for a data-driven interpretation of texts in the humanities.

Inducing Affective Lexical Semantics in Historical Language

The emotional connotation attached to words undergoes language change. In this paper, we aim at estimating the emotion which is associated with a given word in former language stages of English and German. Emotion is represented following the popular Valence-Arousal-Dominance (VAD) annotation scheme. While being more expressive than polarity alone, existing word polarity induction methods are typically not suited for addressing it. To overcome this limitation, we present adaptations of two popular algorithms to VAD. To measure the effectiveness of these methods in diachronic settings, we present two small gold lexicons for historical word emotions which where annotated by scholars with proficiency in the respective language (stage). In contrast to claims in previous work, we find that limiting the training set to words of supposedly stable emotional meaning does in fact decrease performance, rather than boosting it. This contribution thus elaborates algorithmic and experimental results which form the basis of some of our later publications (Buechel et al., 2017; Buechel and Hahn, 2018; Hellrich et al., 2018) yet had to be skipped there, partly for brevity, partly for a change in focus of our work.