Part-of-Speech Tagging on an Endangered Language: a Parallel Griko-Italian Resource

Proceedings of the 27th International Conference on Computational Linguistics, pages 2529–2539 Santa Fe, New Mexico, USA, August 20-26, 2018. 2529 Part-of-Speech Tagging on an Endangered Language: a Parallel Griko-Italian Resource Antonis Anastasopoulos♠ Marika Lekakou♦ Josep Quer♣ Eleni Zimianiti♦ Justin DeBenedetto♠ David Chiang♠ ♠Department of Computer Science and Engineering, University of Notre Dame ♦Department of Philology, University of Ioannina ♣ICREA & Department of Translation and Language Sciences, Universitat Pompeu Fabra aanastas@nd.edu, mlekakou@cc.uoi.gr, josep.quer@upf.edu Abstract Most work on part-of-speech (POS) tagging is focused on high resource languages, or examines low-resource and active learning settings through simulated studies. We evaluate POS tagging techniques on an actual endangered language, Griko. We present a resource that contains 114 nar- ratives in Griko, along with sentence-level translations in Italian, and provides gold annotations for the test set. Based on a previously collected small corpus, we investigate several traditional methods, as well as methods that take advantage of monolingual data or project cross-lingual POS tags. We show that the combination of a semi-supervised method with cross-lingual transfer is more appropriate for this extremely challenging setting, with the best tagger achieving an ac- curacy of 72.9%. With an applied active learning scheme, which we use to collect sentence-level annotations over the test set, we achieve improvements of more than 21 percentage points. 1 Introduction Most natural language processing (NLP) applications have been developed for and tested on only a handful of languages. The majority of the world’s languages are under-represented in the field, mostly due to the lack of proper resources. Endangered languages pose additional problems, as the lack of resources is further exacerbated by the lack of standard orthography. Therefore, there is an obvious need for both resources and technologies adapted to languages of under-represented communities. Especially in the case of endangered languages, computational approaches can be used to scale and accelerate documentation and revitalization efforts. For the purposes of this study, we focus on Griko, a Greek dialect spoken in southern Italy, in the Grec`ıa Salentina area southeast of Lecce.1 There is another endangered Italo-Greek variety in southern Italy spoken in the region of Calabria, known as Grecanico or Greco. Both languages, jointly referred to as Italiot Greek, were included as seriously endangered in the UNESCO Red Book of Endangered Languages in 1999. Griko is only partially intelligible with modern Greek, and unlike other Greek di- alects, it uses the Latin alphabet. Less than 20,000 people (mostly people over 60 years old) are believed to be native speakers (Horrocks, 2009; Douri and De Santis, 2015), a number which is quite likely an overestimation (Chatzikyriakidis, 2010). In general, the lack of annotated resources can be addressed through several directions. The obvious first one is the collection of human annotations, an expensive and time-consuming process. Another option is to collect additional monolingual data and use them in weakly-supervised approaches. Finally, one could use methods that transfer knowledge across languages, in which case they could focus on collecting translations or bilingual data. With this paper we attempt to quantify …

Content truncated to 500 words