Words are Malleable

• DOI: 10.1145/3132847.3132878
• Publication Date: 2017-11-06T13:30:29Z
• Authors: Hosein Azarbonyad; Mostafa Dehghani; Kaspar Beelen; Alexandra Arkut; Maarten Marx; Jaap Kamps
• Keywords: Viewpoints; Semantic change; Similarity (geometry); Impossibility

Recently, researchers started to pay attention the detection of temporal shifts in meaning words. However, most (if not all) these approaches restricted their efforts uncovering change over time, thus neglecting other valuable dimensions such as social or political variability. We propose an approach for detecting semantic between different viewpoints---broadly defined a set texts that share specific metadata feature, which can be time-period, but also entity party. For each viewpoint, we learn space word is represented low dimensional neural embedded vector. The challenge compare one its another and measure size shifts. effectiveness based on optimal transformations two spaces with similarity neighbors respective spaces. Our experiments demonstrate combination performs best. show only occur time along viewpoints short period time. evaluation, how this captures meaningful help improve tasks contrastive viewpoint summarization ideology (measured classification accuracy) texts. laws were empirically shown hold across viewpoints. These state frequent words are less likely shift while many senses more do so.