Grant: Discovery Grants Program - Individual

Wolastoqey is an eastern Algonquian language spoken in regions of New Brunswick, Quebec, and Maine. (This language is also referred to as Maliseet, but many speakers use the term Wolastoqey to refer to their language.) Wolastoqey is endangered, with only roughly 250 people reporting it as their mother tongue in the most recent census. Most speakers are elderly, and most children do not learn the language fluently. Natural language processing (NLP) is a branch of artificial intelligence that focuses on human language, primarily text. Applications of NLP include language technologies such as next word prediction on phones, spelling checkers in word processors, and automatic translation tools. The goal of this research program is to develop language technology to aid the learning and teaching of Wolastoqey. Many contemporary techniques in NLP require a model to be trained on a very large collection of text, referred to as a corpus. Such techniques cannot be directly applied to Wolastoqey because no sufficiently large corpus is available. Furthermore, Wolastoqey, like other eastern Algonqian languages, is polysynthetic, meaning that it has very complex morphology, i.e., word structure. Computational models of morphology are therefore important for building many types of Wolastoqey language technology. There are two Wolastoqey alphabets. One opportunity for language technology identified by Wolastoqiyik is automatic conversion between these alphabets. This would help language teachers to adapt materials from one alphabet to the other and could help to make existing language learning resources available in both alphabets. One challenge to building such a system is that morphological information is required for full coverage. Accessing information in a dictionary can also be a challenge for Wolastoqey learners. Given a wordform, because of the complex morphology, it can often be difficult for a learner to determine which dictionary entry to consult. Again, computational models of morphology could help. This research program will focus on objectives motivated by these challenges. Specifically, this research will consider a finite state model of Wolastoqey inflectional morphology, neural network-based approaches to extending the coverage of this model, unsupervised models of Wolastoqey derivational morphology, and approaches to learning word embeddings for Wolastoqey words. This research will lead to improved methods for automatically converting between Wolastoqey alphabets, improved dictionary access by automatically identifying the headword for a given wordform, reverse dictionary search systems, extended Wolastoqey dictionaries that include information about the structure of stems, and (semi-)automatically constructed, topically-focused glossaries. In the longer term, this research will help to develop a wide range of language technologies for Wolastoqey. This research will also benefit work on other Algonquian and polysynthetic languages.