Semantics-Aware Inferential Network for Natural Language Understanding

Authors: Shuiliang Zhang, Hai Zhao, Junru Zhou, Xi Zhou, Xiang Zhou14437-14445

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiment results indicate that our proposed model achieves significant improvement over the strong baselines on these tasks and obtains the state-of-the-art performance on SNLI and MRQA datasets.
Researcher Affiliation Collaboration Shuailiang Zhang,1,2,3 Hai Zhao,1,2,3 Junru Zhou,1,2,3, Xi Zhou,4 Xiang Zhou4 1Department of Computer Science and Engineering, Shanghai Jiao Tong University 2Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering, Shanghai Jiao Tong University, Shanghai, China 3Mo E Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University, Shanghai, China 4Cloud Walk Technology, Shanghai, China
Pseudocode No The paper includes diagrams and mathematical equations to describe the control, read, and write units (Figures 4, 5, 6) but does not provide structured pseudocode or algorithm blocks.
Open Source Code No Our model framework is based on the Pytorch implementation of transformers2. 2https://github.com/huggingface/transformers. (This refers to a third-party framework used by the authors, not their own source code for the SAIN model.)
Open Datasets Yes We evaluate our model on extractive MRC such as SQu AD (Rajpurkar, Jia, and Liang 2018) and MRQA1 (Fisch et al. 2019) where the answer is a span of the passage. MRQA is a collection of existing question-answering related MRC datasets, such as Search QA (Dunn et al. 2017), News QA (Trischler et al. 2017), Natural Questions (Kwiatkowski et al. 2019), Trivia QA (Joshi et al. 2017), etc. Natural Language Inference ... We evaluate on 4 diverse datasets, including SNLI (Bowman et al. 2015), MNLI (Williams, Nangia, and Bowman 2018), QNLI (Rajpurkar et al. 2016) and RTE (Bentivogli et al. 2009).
Dataset Splits Yes Ablation study on the RTE and SQu AD dev sets as shown in Table 3. To the best of our knowledge, we achieve state-of-theart performance on MRQA (dev sets) and SNLI.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments.
Software Dependencies No Our model framework is based on the Pytorch implementation of transformers2. To obtain the semantic role labels, we use the SRL system of (He et al. 2017) as implemented in Allen NLP (Gardner et al. 2018). (The paper mentions software components like PyTorch, Transformers, and AllenNLP but does not provide specific version numbers for these dependencies.)
Experiment Setup Yes According to the experimental results, it is a reasonable configuration that sets the maximum number of predicateargument structures (reasoning steps) M to 3 and 4 for MRC and NLI tasks, respectively. Our model framework is based on the Pytorch implementation of transformers2. We use Adam as our optimizer with initial learning rate 1e-5 and warm-up rate of 0.1. The batch size is set to 8.