What can be learned about a language’s morphology just by observing a list of words? In this talk, I will investigate this question leveraging concepts from NLP. Given a list of words, the computer is tasked with segmenting each word in a wordlist automatically, and the output is evaluated by comparing to human-annotated segmentations. This turns out to be a very challenging task due to the fact that different languages may have completely different morphological structures. I will describe a specific system we designed to achieve this task, and particularly what information / evidence we have used in terms of language typology and language universals. Experments are done in several languages and show that the system can do a resonable job on learning language morphology.

 

Location: 3401 Walnut Street, Room 401B