Morphology-rich Alphasyllabary Embeddings

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2590–2595 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2590 Morphology-rich Alphasyllabary Embeddings Amanuel Mersha, Stephen Wu Addis Ababa University, UT Health Science Center at Houston amanuel.negash@aau.edu.et, wu.stephen.t@gmail.com Abstract Word embeddings have been successfully trained in many languages. However, both intrinsic and extrinsic metrics are variable across languages, especially for languages that depart significantly from English in morphology and orthography. This study focuses on building a word embedding model suitable for the Semitic language of Amharic (Ethiopia), which is both morphologically rich and written as an alphasyllabary (abugida) rather than an alphabet. We compare embeddings from tailored neural models, simple pre-processing steps, off-the-shelf baselines, and parallel tasks on a better-resourced Semitic language – Arabic. Experiments show our model’s performance on word analogy tasks, illustrating the divergent objectives of morphological vs. semantic analogies. Keywords: word embeddings, word2vec, Amharic, morphologically rich, alphasyllabary 1. Introduction Word embeddings are ubiquitous in today’s language- related applications. A long history of distributional semantics efforts (Deerwester et al., 1990; Blei et al., 2003; Padó and Lapata, 2007) have largely given way to efficient neural network-inspired word embedding methods (Mikolov et al., 2013; Pennington et al., 2014), which have been applied across dozens of tasks with great success. Word embeddings have also been studied in bilingual (Zou et al., 2013) and multilingual (Ammar et al., 2016; Bojanowski et al., 2017) settings, but evaluations in lesser-resourced languages have been cursory and highly variable. This work focuses on word embeddings in the Amharic language, the second-most-spoken Semitic language1 (Eberhard et al., 2020). Similar to other Semitic lan- guages like Arabic or Hebrew, it is morphologically rich, with templatic morphology that is fusional rather than agglutinative, so that morphemes are difficult to separate: አይመለስም almelesem return.1sg.neg.past እመለሳለሁ imelesalehu return.1sg.future Amharic is written with the Fidel alphasyllabary (also known as an abugida). Each character in Amharic Fi- del represents both a consonant (35) and a vowel (7 + 5 diphthongs) and is unique, yet there are similarities for each onset (consonant) and for each nucleus (vowel) of syllables. Below, note the similarities in shape cor- responding to consonants as well as vowels: Note that ሀ ሁ ሂ ሃ ሄ ህ ሆ ha hu hi ha hē hə ho ለ ሉ ሊ ላ ሌ ል ሎ le lu li la lē lə lo spoken Amharic does not follow the syllabic structure 1https://www.ethnologue.com/language/amh of the Fidel characters strictly; for example, the 6th form ￿ may have an ə] vowel or no vowel at all. As we will show, word embedding models face unique challenges in this lesser-studied language. An abugida orthographic system as described above (besides Fi- del, see also e.g., Devanagari) is distinct from an abjad (requisite glyphs only represent consonants, e.g., Ara- bic), syllabary (glyphs represent syllables with no vi- sual similarity between related sounds, e.g., Japanese hiragana or katakana), or alphabet (glyphs give vowels equal status to consonants). Thus, this work introduces both …

Content truncated to 500 words