Sentence Completion using NLP Techniques

Umang Rupareliya
2019 International Journal for Research in Applied Science and Engineering Technology  
The paper aims to study, analyse and compare the results of sentence completion task based on SAT style questions. In this paper, we plan to apply and compare various approaches for automated sentence completion which include Latent Semantic Indexing and Recurrent Neural Networks. Results from the past research stated that the LSA model outperforms the conventional n-gram model models on the Microsoft Research Sentence Completion Challenge. Hence, we will analyse and compare the results of the
more » ... SA model with RNN model to decide which of them performs better considering the scale of the data. The tasks involve training on a large corpus of unannotated text, to then try to predict the missing words in the test set which contains thousands of sentences where one word is missing and five alternatives for the missing word. Methods using local information and global information for the task of sentence completion are used and we find that method using global information (Latent Semantic Analysis) proves to be better than the method using local information (Recurrent Neural Network). We compare our approach to Microsoft research sentence completion challenge by extending RNN to LSTM with RNN.
doi:10.22214/ijraset.2019.4474 fatcat:hslskodrf5dppagxj44kk6ynma