Automatic cognate identification with gap-weighted string subsequences.

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1227–1231, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Automatic cognate identification with gap-weighted string subsequences. Taraka Rama Spr˚akbanken University of Gothenburg Box 200, Gothenburg, Sweden taraka.rama.kasicheyanula@gu.se Abstract In this paper, we describe the problem of cog- nate identification in NLP. We introduce the idea of gap-weighted subsequences for dis- criminating cognates from non-cognates. We also propose a scheme to integrate phonetic features into the feature vectors for cognate identification. We show that subsequence based features perform better than state-of- the-art classifier for the purpose of cognate identification. The contribution of this paper is the use of subsequence features for cognate identification. 1 Introduction Cognates are words across languages whose ori- gin can be traced back to a common ancestral lan- guage. For example, English ∼ German night ∼ Nacht ‘night’ and English hound ∼ German Hund ‘dog’ are cognates whose origin can be traced back to Proto-Germanic. Sometimes, cognates are not revealingly similar but have changed substantially over time such that they do not share form simi- larity. An example of such a cognate pair is the English wheel and Sanskrit chakra ‘wheel’, which can be traced back to Proto-Indo-European (PIE) ∗kwekwelo. Automatic cognate identification, in NLP, refers to the application of string similarity or phonetic similarity algorithms either independently, or in tan- dem with machine learning algorithms for determin- ing if a given word pair is cognate or not (Inkpen et al., 2005). In NLP, even borrowed words (loan- words) are referred to as cognates. In contrast, his- torical linguistics makes a stark distinction between loanwords and cognates. For example, English beef is a loanword from Norman French. In this paper, we use cognates to refer to those words whose origin can be traced back to a com- mon ancestor. We use string subsequence based fea- tures (motivated from string kernels) for automatic cognate identification. We show that subsequence- based features outperform word similarity measures at the task of automatic cognate identification. We motivate the use of subsequence based features in terms of linguistic examples and then proceed to formulate the subsequence based features that can be derived from string kernels (Shawe-Taylor and Cristianini, 2004). In information retrieval litera- ture, string subsequences go under the name of skip- grams (J¨arvelin et al., 2007). 2 Related work The approaches developed by Kondrak and Sherif (2006) and Inkpen et al. (2005) supply different string distances between a pair of words as features to a linear classifier. Usually, a linear classifier such as SVM is trained on labeled positive (“cognates”) and negative (“non-cognates”) examples and tested on a held-out dataset. Basic vocabulary lists such as the ones devised by Morris Swadesh (Swadesh, 1952), provide a suitable testing ground for apply- ing machine learning algorithms to automatically identify cognates. Some standardized word lists come with cognate information and, subsequently, can used to infer the relationship between the lan- guages (Dyen et …

Content truncated to 500 words