Kvistur 2.0: a BiLSTM Compound Splitter for Icelandic

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3991–3995 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3991 Kvistur 2.0: a BiLSTM Compound Splitter for Icelandic Jón Friðrik Daðason1, David Erik Mollberg1, Hrafn Loftsson1, Kristín Bjarnadóttir2 1Department of Computer Science, 2The Árni Magnússon Institute for Icelandic Studies 1Reykjavik University, 2University of Iceland 1{jond19, de14, hrafn}@ru.is, 2kristinb@hi.is Abstract In this paper, we present a character-based BiLSTM model for splitting Icelandic compound words, and show how varying amounts of training data affects the performance of the model. Compounding is highly productive in Icelandic, and new compounds are constantly being created. This results in a large number of out-of-vocabulary (OOV) words, negatively impacting the performance of many NLP tools. Our model is trained on a dataset of 2.9 million unique word forms and their constituent structures from the Database of Icelandic Morphology. The model learns how to split compound words into two parts and can be used to derive the constituent structure of any word form. Knowing the constituent structure of a word form makes it possible to generate the optimal split for a given task, e.g., a full split for subword tokenization, or, in the case of part-of-speech tagging, splitting an OOV word until the largest known morphological head is found. The model outperforms other previously published methods when evaluated on a corpus of manually split word forms. This method has been integrated into Kvistur, an Icelandic compound word analyzer. Keywords: compound splitting, morphology, BiLSTM 1. Introduction Compounds are extremely common in Icelandic, account- ing for over 88% of all words in the Database of Icelandic Morphology (DIM) (Bjarnadóttir, 2017; Bjarnadóttir et al., 2019). As compounding is so productive, new compounds frequently occur as out-of-vocabulary (OOV) words, which may adversely affect the performance of NLP tools. Fur- thermore, Icelandic is a morphologically rich language with a complex inflectional system. There are 16 inflectional cat- egories (i.e., word forms with unique part-of-speech (PoS) tags) for nouns, for adjectives 120, and for verbs 122, ex- cluding impersonal constructions. The average number of inflectional forms per headword in DIM is 21.7. Included in this average are all uninflected words as well inflectional variants, i.e., dual word forms with the same PoS tag. Compounds are formed by combining two words, which may be compounds themselves. The former word is known as a modifier and the second as a head, assuming binary branching (Bjarnadóttir, 2005). Theoretically, there is no limit to how many constituents a compound can be com- posed of, although very long words such as uppáhald- seldhúsinnréttingaverslunin ‘the favorite kitchen furniture store’ (containing 7 constituent parts) are rare. The con- stituent structure of a compound word can be represented by a full binary tree, as shown in Figure 1. Compound splitting, or decompounding, is the process of breaking compound words into their constituent parts. This can significantly reduce the number of OOV words for languages where compounding is productive. Compound splitting has been shown to …

Content truncated to 500 words