Using lexical language models to detect borrowings in monolingual wordlists.

Area of Science:

• Computational Linguistics
• Language Evolution
• Phonology

Background:

• Lexical borrowing is a frequent process in language evolution.
• Automated detection of borrowings is challenging, often overlooking expert linguistic strategies.
• Phonological and phonotactic clues are vital for identifying recent, unadapted borrowings.

Purpose of the Study:

• To test the effectiveness of phonological and phonotactic clues in automated lexical borrowing detection.
• To propose a framework for supervised borrowing detection in monolingual wordlists.
• To investigate the performance of machine learning models in identifying borrowings.

Main Methods:

• Modeling phonology and phonotactics using Support Vector Machines, Markov models, and recurrent neural networks.
• Developing a supervised framework for borrowing detection in monolingual wordlists.
• Utilizing a revised dataset annotated for lexical borrowings across 41 diverse languages.

Main Results:

• Initial results for monolingual borrowing detection were largely unsatisfying.
• Model performance improved with increased amounts of attested borrowings.
• Performance was better when borrowings originated predominantly from a single donor language.

Conclusions:

• Phonological and phonotactic clues from monolingual data alone are often insufficient for robust borrowing detection.
• These clues show promise for integrated approaches incorporating multilingual information.
• Further research is needed to enhance automated lexical borrowing detection systems.