Automatic Speech Recognition and Translation for Low Resource Languages

7Syllable-Level Morphological Segmentation of Kannada and Tulu Words

Asha Hegde* and Hosahalli Lakshmaiah Shashirekha^†

Department of Computer Science, Mangalore University, Mangalore, India

Abstract

Morphological segmentation (MS) is the study of segmentation of words into the smallest units called morphemes. It is performed as an additional preprocessing step for many natural language processing (NLP) downstream tasks, such as machine translation, text classification, and information retrieval. Morphological segmentation of words in agglutinative languages is critical as the words are formed by stringing of suffixes. Further, the richness in morphology, inflections, and sandhi mechanisms complicate the segmentation task. The MS task is achieved by the detection of morpheme boundaries (MB) in the compound words which, in turn, produces the root word and the suffixes and/or prefixes. Determining the MB is a preliminary step in MS and rule-based, unsupervised, or supervised learning approaches are used to determine the MB. The inefficiency of rule-based and unsupervised approaches for MS has led to the development of supervised models for MS. High-resource languages such as English, Spanish, and German have witnessed significant research efforts for MS. However, this task is rarely explored in under-resourced agglutinative languages, such as Kannada, Tamil, Telugu, Tulu, and Malayalam. To address this gap, this chapter describes the construction of annotated Kannada and Tulu ...