DialectGram uncovers hidden linguistic diversity across cities, states, and countries!
A new model called DialectGram has been developed to detect dialectal variation at different levels of geographic resolution without needing to pre-define regions. DialectGram learns dialect-sensitive word meanings and can analyze dialectal differences at any chosen resolution after training. It can estimate the proportion of different word meanings used in a region and does not require re-training when regions change. DialectGram was tested against other models on a new evaluation dataset called DialectSim and was found to effectively model linguistic variation.
