/DeepLexicalHypothesis

Primary LanguageJupyter NotebookMIT LicenseMIT

DeepLexicalHypothesis

Reproduces experiments in the Lexical Hypothesis chapter of Language Modeling for Personality Prediction. Navigate to the Jupyter Notebook (DeepLexicalHypothesis.ipynb) and press play. Runs on Google's colab servers.

The lexical hypothesis claims that natural language includes all the distinctions we find worthwhile to make about one another; in order to understand the structure of personality we should look to language. Big Five, the dominant model in the field, claims to be the first five principal components of character descriptions in English. The code here is an alternative method to vectorize words with deep learning models. Results show just two factors: socialization and self-actualization. This is interesting because a search for a descriptive framework produces something that aligns neatly with theories of personality development. It also supports Digman's empirical work who found a similar two traits in a meta-analysis of Big Five survey data.