Neural Network Models for Paraphrase Identification, Semantic Textual Similarity, Natural Language Inference, and Question Answering
This provides a comparative benchmark for researchers in NLP tasks like paraphrase identification and question answering, though it is incremental as it evaluates existing methods.
The paper systematically analyzes neural network designs for sentence pair modeling across eight datasets, finding that LSTM-based contextual encoding and inter-sentence interactions are critical, with specific models performing best depending on dataset size.
In this paper, we analyze several neural network designs (and their variations) for sentence pair modeling and compare their performance extensively across eight datasets, including paraphrase identification, semantic textual similarity, natural language inference, and question answering tasks. Although most of these models have claimed state-of-the-art performance, the original papers often reported on only one or two selected datasets. We provide a systematic study and show that (i) encoding contextual information by LSTM and inter-sentence interactions are critical, (ii) Tree-LSTM does not help as much as previously claimed but surprisingly improves performance on Twitter datasets, (iii) the Enhanced Sequential Inference Model is the best so far for larger datasets, while the Pairwise Word Interaction Model achieves the best performance when less data is available. We release our implementations as an open-source toolkit.