AmritaCEN at SemEval-2016 Task 11: Complex Word Identification using Word Embedding

Proceedings of SemEval-2016, pages 1022–1027, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics AmritaCEN at SemEval-2016 Task 11: Complex Word Identification using Word Embedding Sanjay S.P and Anand Kumar M and Soman K P Centre for Computational Engineering & Networking (CEN) Amrita School of Engineering, Coimbatore Amrita Vishwa Vidyapeetham Amrita University India sanjay.poongs@gmail.com, m anandkumar@cb.amrita.edu, kp soman@amrita.edu Abstract Complex word identification task focuses on identifying the difficult word from English sentence for a Non-Native speakers. Non- Native speakers are those who don’t have En- glish as their native language. It is a sub- task for lexical simplification. We have ex- perimented with word embedding features, or- thographic word features, similarity features and POS tag features which improves the per- formance of the classification. In addition to the SemEval 2016 results we have evaluated the training data by varying the vector dimen- sion size and obtained the best possible size for producing better performance. The SVM learning algorithm will attains constant and maximum accuracy through linear classifier. We achieve a G-score of 0.43 and 0.54 on computing complex words for two systems. 1 Introduction Complex Word Identification (CWI) is the task of identifying difficult words for a Non-Native speaker. The main objective of Complex Word Identification is to simplify and enhance the perplexing words in the sentences for Non-Native speakers. Complex Word Identification is used in Lexical Simplifica- tion (LS). The Lexical Simplification is important for all Text Simplification (Gustavo Henrique Paet- zold, 2006). The perspective of each user is dif- ferent therefore the task Complex Word Identifica- tion have been demanding one. For distinguish com- plex words for individual user will be a tedious task, therefore the CWI task is a difficult one, so we are identifying the complex words for a group of users. CWI could be used in summarization of sto- ries or generate abstract of a book. Shortening a paragraph would be difficult without removing the complex words, since the shortened text with complex words would be very difficult or confus- ing for non-native users to read and understand. We can make a story more simple for kids to read by replacing complex words with most common words (Tomoyuki Kajiwara et al., 2013). The main ways to identify the complex words are Threshold- based and Classification-based approches (Gustavo Henrique Paetzold, 2006). In Threshold-based ap- proach, length, number of syllables, count of ambi- guity and word frequency are considered as features. If the values are above the threshold, then the word is said to be a complex word. In the Classifier-based approach, the word features are trained and the word features which are similar to the trained word fea- tures are considered to be a complex word. 2 Complex Word Identification (CWI) - Task Description The SemEval-2016 organizers have released three files cwi training, cwi training allannotations, and cwi testing. The first two files are training files and the third file is a testing file. Our task is to find the words in the testing …

Content truncated to 500 words