Abstract:

Semantic processing of morphologically rich languages is still at its infancy and is often portrayed as a daunting task. The Inflection of stems with multiple discontiguous morphemes that results in high variations of the surface form, flexibility in free word order sentence formations, and high productivity of the independent words in such languages calls for new methods for processing such constructs. In this talk, I will focus on two
tasks, segmentation, and identification of noun pairs related through derivational morphology. Below is a brief sketch of what I will discuss under these tasks.

Segmentation – In Sanskrit texts, words are often joined together making phoneme level changes at word boundaries. We treat the problem as a query expansion problem under path constrained random walks framework. Given all syntactically valid segmentations, we aim at finding the most semantically valid combination of words to obtain a segmented sentence. We consider the sentence to be a heterogenous information network, and perform multiple random walks over paths that are formed as linguistically motivated
Inductive logic formulations. We approximate the semantic compatibility from distributional information of the words obtained using hand-crafted morphological constraints.

Identification of Noun Pairs related through Derivational Morphology – The semantic sense an affix carries in case of derivational nouns have been much debated in linguistics. In Sanskrit, derivational nouns formed by
affixation of Taddhita affixes are highly productive in nature. Identifying the pedagogical information of such nouns help us in maintaining richer lexicons. The semantic senses and affixes follow a many to many mapping, thereby raising a computational challenge. In our work, we propose a 3 step transductive learning approach to identify pairs of source and derived nouns. We successfully integrate rules from Panini’s Ashtadhyayi, lexical information from word pairs matching the patterns, and distributional information obtained from english synonym word vectors.

About the Speaker:

Pawan Goyal received his B. Tech. degree in Electrical Engineering from IIT Kanpur in 2007 and his Ph.D. degree in the faculty of Computing and Engineering from University of Ulster, UK in 2011. He was then a Post Doctoral Fellow at INRIA Paris Rocquencourt. Currently, he is an Assistant Professor at the Department of Computer Science and Engineering, IIT Kharagpur. His research interests include Natural Language Processing,
Text Mining, Summarization and Sanskrit Computational Linguistics.