Negativity and Semantic Change - Will Hamilton, Stanford University

Explore the concept of negative differentiation in natural language through a comprehensive lecture by Stanford University's Will Hamilton. Delve into the diachronic linguistic mechanisms associated with this phenomenon and learn how dynamic word embeddings are used to test the semantic stability of negative lexical items compared to positive ones. Discover preliminary findings suggesting faster rates of semantic change for negative affectual language. Examine the practical implications of this positive/negative asymmetry for modern sentiment analysis tools. Gain insights into Hamilton's research background, including his work at Stanford University and previous studies at McGill University. Follow the lecture's structure, covering topics such as the linguistic positivity/negativity bias, quantitative approaches to semantic change, word embedding techniques, statistical models of semantic change, and a case study on the word "hate". Enhance your understanding of natural language processing, sentiment analysis, and the evolving nature of language.

Intro
The linguistic positivity/negativity bias
Negative language is more "differentiated"
Negative language is more complex
Psychological factors
Diachronic negative differentiation
A quantitative approach to semantic change
Word embeddings (the basic idea)
Diachronic word embeddings
Different embedding approaches
Sanity check: Do the embeddings capture semantic change?
Which embeddings work best? (3)
Statistical models semantic change: basic trends
Quantifying the association between negativity and semantic change
Accounting for shifting sentiment
Experimental setup and dataset
Preliminary results
Next steps
Sentiment analysis: Some background
Off-the-shelf sentiment analysis
Hypothesis
Lexicons
A case-study in "hate" (examples)
Summary of part 2