Classifying evolutionary forces in language change using neural networks

Area of Science:

• Computational linguistics
• Evolutionary biology
• Cultural evolution

Background:

• Distinguishing neutral evolution (stochastic drift) from selection pressures is a key challenge in studying language and cultural change.
• Existing methods, like the Frequency Increment Test, face limitations in accurately identifying these evolutionary forces.

Purpose of the Study:

• To introduce a novel deep neural network-based technique for detecting evolutionary forces in cultural change.
• To reformulate the detection of evolutionary forces as a binary classification task for improved accuracy.

Main Methods:

• Utilized deep neural networks, specifically residual networks for time series analysis.
• Trained the model on artificially generated samples of cultural change data.
• Compared the neural network model's performance against the Frequency Increment Test.

Main Results:

• The deep neural network technique efficiently and accurately learned to distinguish between stochastic drift and selection pressures.
• The model demonstrated consistent performance in identifying distinctive time series aspects for each evolutionary process.
• The neural time series classification system addressed key limitations of the Frequency Increment Test.

Conclusions:

• Deep neural networks offer a powerful and accurate method for analyzing evolutionary forces in cultural and linguistic change.
• This new technique provides a robust solution for differentiating neutral drift from selection-driven changes.
• The study highlights the potential of machine learning in advancing the understanding of cultural evolution.