Question Answering Using Regularized Match-LSTM and Answer Pointer
نویسنده
چکیده
Automated reading comprehension is an important problem in natural language processing. The Stanford Question Answering Dataset (SQuAD) is a convenient set of questions and crowdsourced answers to utilize for evaluation of QA systems. In this paper, we implement a version of an architecture proposed by Wang and Jiang (2016) based on match-LSTM [9], a model used for textual entailment, and answer candidate generation based on Pointer Net (Vinyals et al., 2015) [7]. We extend the original implementation with an investigation of the effect of regularization methods on this task. Specifically, we argue that any amount of regularization via dropout improves test performance because it prevents overfitting; however, we also note that varying the dropout probability does not significantly change performance, provided it is sufficiently far from 1 (< 0.8).
منابع مشابه
Improving Match-LSTM for Machine Comprehension
Machine comprehension is a critical problem that lies on the frontier of natural language processing. The Stanford Question Answering Dataset (SQuAD), offers a set of questions and their answers created by humans through crowdsourcing. We implemented an end-to-end neural architecture for the task based on MatchLSTM and Pointer Net, inspired by previous work done by Wang and Jiang in Machine Com...
متن کاملCoattention Answer-Pointer Networks for Question Answering
Machine comprehension (MC) and question answering (QA) are crucial tasks in natural language understanding. Training deep neural network-based QA models has become practical upon the recent release of the Stanford Question Answering Dataset (SQuAD), a significantly larger dataset of question-answer pairs created by humans on a set of Wikipedia articles [1]. In this paper, we propose an end-to-e...
متن کاملMachine Comprehension Using Match-LSTM and Answer Pointer
Machine comprehension of text is an important problem in natural language processing. A recently released dataset, the Stanford Question Answering Dataset (SQuAD), offers a large number of real questions and their answers created by humans through crowdsourcing. SQuAD provides a challenging testbed for evaluating machine comprehension algorithms, partly because compared with previous datasets, ...
متن کاملQuestion Answering Using Match-LSTM and Answer Pointer
Machine comprehension of text is a significant problem in natural language processing today – in this project, we tackle machine reading comprehension as applied to question answering. Our goal is: given a question and a context paragraph, to extract from the paragraph the answer to the question. As an oracle, on the dataset we used, humans score over 86.8% accuracy (EM) on the test set for thi...
متن کاملSing M Atch - Lstm and a Nswer P Ointer
Machine comprehension of text is an important problem in natural language processing. A recently released dataset, the Stanford Question Answering Dataset (SQuAD), offers a large number of real questions and their answers created by humans through crowdsourcing. SQuAD provides a challenging testbed for evaluating machine comprehension algorithms, partly because compared with previous datasets, ...
متن کامل