Constraint-Based Models of Lexical Borrowing

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 598–608, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Constraint-Based Models of Lexical Borrowing Yulia Tsvetkov Waleed Ammar Chris Dyer Language Technologies Institute Carnegie Mellon University Pittsburgh, PA, 15213, USA {ytsvetko, wammar, cdyer}@cs.cmu.edu Abstract Linguistic borrowing is the phenomenon of transferring linguistic constructions (lexical, phonological, morphological, and syntactic) from a “donor” language to a “recipient” lan- guage as a result of contacts between commu- nities speaking different languages. Borrowed words are found in all languages, and—in con- trast to cognate relationships—borrowing rela- tionships may exist across unrelated languages (for example, about 40% of Swahili’s vocabu- lary is borrowed from Arabic). In this paper, we develop a model of morpho-phonological transformations across languages with features based on universal constraints from Optimality Theory (OT). Compared to several standard— but linguistically naïve—baselines, our OT- inspired model obtains good performance with only a few dozen training examples, making this a cost-effective strategy for sharing lexical information across languages. 1 Introduction We may imagine that globalization is a modern phe- nomenon, but the lexicons of the world’s languages attest to the fact that robust interaction between communities of speakers of different languages is widespread throughout history. Language contact breeds linguistic borrowing—a phenomenon as old as language itself—adoption and nativization of phonemes, morphemes, words, and syntactic con- structions from another language (Thomason and Kaufman, 2001). Contact-induced borrowing is a fundamental re- search topic in linguistics; however, in computational linguistics, very limited work has addressed model- ing this phenomenon. The problem we address is the identification of plausible donor words (in the donor language) given a loanword (in the recipient language), and vice versa, identification of loanwords given a donor. For example, given a Swahili loan- word safari ‘journey’, our model identifies its Arabic donor ������� (sfryh)1 ‘journey’ (§2). Although at a high level, this is an instance of the well-known prob- lem of modeling string transductions, our interest is being able to identify correspondences across lan- guages with minimal supervision, so as to make the technique applicable in low-resource settings. To re- duce the supervision burden, we propose a model that includes awareness of the morpho-phonological re- pair strategies that native speakers of a language sub- consciously employ to adapt a loanword to phonolog- ical constraints of the recipient language (§3). To this end, we use constraint-based theories of phonology, as exemplified by Optimality Theory (OT) (Prince and Smolensky, 2008; McCarthy, 2009), which non- computational linguistic work has demonstrated to be particularly well suited to account for phonolog- ically complex borrowing processes (Kang, 2011). We operationalize OT constraints as features in our borrowing model (§4). We conduct a case study on Arabic and Swahili, two unrelated languages with a long history of con- tact; we then apply the model to additional language pairs (§5). The proposed approach significantly out- performs transliteration and cognate discovery mod- els (§6). 1We use Buckwalter notation to write Arabic glosses. …

Content truncated to 500 words