Statistical structure and the evolution of languages.

Area of Science:

• Linguistics
• Computational Social Science
• Natural Language Processing

Background:

• Human cultural development is characterized by the emergence of new words and ideas.
• Understanding the mechanisms driving language evolution is crucial for comprehending societal change.

Purpose of the Study:

• To investigate the statistical traces of cultural development using modern natural language processing techniques.
• To compare different models of language evolution and identify previously unrecognized regularities.

Main Methods:

• Utilized word embeddings, a natural language processing method, to analyze semantic and temporal patterns in language.
• Examined embeddings from English and 21 other languages to identify statistical regularities.

Main Results:

• Identified four key regularities: frequency assortativity, characteristic clustering velocity profiles, persistent temporal dynamics, and Taylor's law.
• Demonstrated that these regularities systematically quantify historical fluctuations in neologisms.
• Proposed a generative model (directed preferential placement) that replicates these observed regularities.

Conclusions:

• The identified regularities provide a quantitative framework for understanding language and cultural evolution.
• The proposed generative models offer insights into the underlying processes of semantic and cultural change.
• Analogous regularities in other datasets suggest broader applicability of these findings to cultural dynamics.