Context Sensitive Lemmatization Using Two Successive Bidirectional Gated Recurrent Networks

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1481–1491 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1136 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1481–1491 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1136 Context Sensitive Lemmatization Using Two Successive Bidirectional Gated Recurrent Networks Abhisek Chakrabarty Onkar Arun Pandit Computer Vision and Pattern Recognition Unit Indian Statistical Institute 203 B.T. Road, Kolkata-700108, India abhisek0842@gmail.com, oapandit@gmail.com, utpal@isical.ac.in Utpal Garain Abstract We introduce a composite deep neural network architecture for supervised and language independent context sensitive lemmatization. The proposed method con- siders the task as to identify the correct edit tree representing the transformation between a word-lemma pair. To find the lemma of a surface word, we exploit two successive bidirectional gated recur- rent structures - the first one is used to ex- tract the character level dependencies and the next one captures the contextual infor- mation of the given word. The key ad- vantages of our model compared to the state-of-the-art lemmatizers such as Lem- ming and Morfette are - (i) it is indepen- dent of human decided features (ii) except the gold lemma, no other expensive mor- phological attribute is required for joint learning. We evaluate the lemmatizer on nine languages - Bengali, Catalan, Dutch, Hindi, Hungarian, Italian, Latin, Roma- nian and Spanish. It is found that ex- cept Bengali, the proposed method outper- forms Lemming and Morfette on the other languages. To train the model on Ben- gali, we develop a gold lemma annotated dataset1 (having 1, 702 sentences with a total of 20, 257 word tokens), which is an additional contribution of this work. 1 Introduction Lemmatization is the process to determine the root/dictionary form of a surface word. Morpho- logically rich languages suffer due to the existence 1The dataset and the code of model architecture are re- leased with the paper. They are also available in http: //www.isical.ac.in/˜utpal/resources.php of various inflectional and derivational variations of a root depending on several linguistic proper- ties such as honorificity, parts of speech (POS), person, tense etc. Lemmas map the related word forms to lexical resources thus identifying them as the members of the same group and providing their semantic and syntactic information. Stem- ming is a way similar to lemmatization producing the common portion of variants but it has several limitations - (i) there is no guarantee of a stem to be a legitimate word form (ii) words are consid- ered in isolation. Hence, for context sensitive lan- guages i.e. where same inflected word form may come from different sources and can only be dis- ambiguated by considering its neighbouring infor- mation, there lemmatization defines the foremost task to handle diverse text processing problems (e.g. sense disambiguation, parsing, translation). The key contributions of this work are as fol- lows. We address context sensitive lemmatiza- tion introducing a two-stage bidirectional gated recurrent neural network (BGRNN) architecture. Our model is …

Content truncated to 500 words