Languages with more speakers tend to be harder to (machine-)learn

Area of Science:

• Cognitive Science
• Computational Linguistics
• Language Acquisition

Background:

• Language models (LMs) offer a computational framework for studying human language.
• Previous research suggested languages with larger speaker populations are easier to learn.

Purpose of the Study:

• To test the hypothesis that languages with more speakers are easier to learn using LMs.
• To empirically investigate the relationship between speaker population size and language learning difficulty.

Main Methods:

• Trained various LMs, from n-gram to deep neural networks, on a corpus of 1293 languages.
• Employed quantitative methods and machine learning to estimate learning difficulty.
• Accounted for language relatedness (phylogenetic) and geographical proximity.

Main Results:

• Found a robust relationship between language learning difficulty and speaker population size.
• Contrary to expectations, results indicate languages with more speakers tend to be harder to learn.
• This finding challenges previous assumptions in linguistic research.

Conclusions:

• Language learning difficulty is inversely correlated with speaker population size.
• LMs provide a valuable tool for exploring linguistic hypotheses.
• The study necessitates a re-evaluation of factors influencing language learnability.