Building a Constraint Grammar Parser for Plains Cree Verbs and Arguments

Building a Constraint Grammar Parser for Plains Cree Verbs and Arguments Katherine Schmirler1, Antti Arppe1, Trond Trosterud2, Lene Antonsen2 1University of Alberta, 2UIT Arctic University of Norway {schmirle, arppe}@ualberta.ca, {trond.trosterud, lene.antonsen}@uit.no Abstract This paper discusses the development and application of a Constraint Grammar parser for the Plains Cree language. The focus of this parser is the identification of relationships between verbs and arguments. The rich morphology and non-configurational syntax of Plains Cree make it an excellent candidate for the application of a Constraint Grammar parser, which is comprised of sets of constraints with two aims: 1) the disambiguation of ambiguous word forms, and 2) the mapping of syntactic relationships between word forms on the basis of morphological features and sentential context. Syntactic modelling of verb and argument relationships in Plains Cree is demonstrated to be a straightforward process, though various semantic and pragmatic features should improve the current parser considerably. When applied to even a relatively small corpus of Plains Cree, the Constraint Grammar parser allows for the identification of common word order patterns and for relationships between word order and information structure to become apparent. Keywords: Plains Cree, Constraint Grammar, automatic parser 1. Introduction This paper presents the first developmental stage of a Constraint Grammar (CG) parser for Plains Cree, a North American Indigenous language spoken by several thousands of people primarily in the Canadian provinces of Alberta and Saskatchewan. It is a member of the Algonquian language family and the westernmost language of the Cree-Montagnais-Naskapi continuum spoken across much of Canada. A number of texts have been published in Plains Cree, totalling several hundreds of thousands of words; while these comprise a small corpus compared to many available for languages such as English, they do allow for the development and testing of computational tools such as morphological and syntactic models. Once such tools have been developed, they can be used not only for corpus investigations, but can be implemented in various language technological applications for speakers, teachers, and students of Plains Cree. These include morphologically “intelligent” online dictionaries, morphosyntactically tagged corpora, spell checkers, grammar checkers, and intelligent computer-assisted language learning (ICALL) applications. 1.1 Research Questions The present research aims to answer three questions. First, to what extent can basic syntactic relationships in Plains Cree be adequately modelled using only morphological information? Second, of those syntactic relationships that cannot be identified, how can they be accounted for in future development? Third, when applied to a corpus of Plains Cree, what word order patterns can be identified? 2. Plains Cree Morphosyntax Like all Algonquian languages, Plains Cree is highly polysynthetic with a complex morphological system. The morphosyntactic features and inflectional system of Plains Cree are modelled by means of a morphological analyser (Snoek et al, 2014; Harrigan et al., 2017), the output of which includes morphosyntactic feature tags and all possible analyses of ambiguous forms. These analyses become the input for the Constraint Grammar parser. The 1 Circumflexes or macrons are used to mark long vowels in written Plains Cree; …

Content truncated to 500 words