Proposed a multimodal one-shot learning method using Siamese CNN, doubling accuracy in 11-way cross-modal matching.

Ryan Eloff, Herman A. Engelbrecht, Herman Kamper

Proposed iterative beam search algorithm enhances dialogue generation diversity and quality.

Ilia Kulikov, Alexander H. Miller, Kyunghyun Cho et al.

ReCoRD uses automatically generated news-based dataset emphasizing commonsense reasoning, with models scoring far below human performance.

Sheng Zhang, Xiaodong Liu, Jingjing Liu et al.

Hierarchical semantic parsing with RNNG outperforms Seq2Seq on 44k annotated dialog requests, enabling complex nested query understanding.

Sonal Gupta, Rushin Shah, Mrinal Mohit et al.

Introduces a Chinese span-extraction MRC dataset with nearly 20,000 questions, including a challenging set requiring multi-sentence reasoning.

Yiming Cui, Ting Liu, Wanxiang Che et al.

Proposed a deterministic mapping from UD features to UniMorph, achieving 64.13% macro-average recall.

Arya D. McCarthy, Miikka Silfverberg, Ryan Cotterell et al.

MELD extends EmotionLines with multimodal, multi-party emotion recognition, offering 13,000 annotated utterances.

Soujanya Poria, Devamanyu Hazarika, Navonil Majumder et al.

Introduced OpenBookQA dataset with 1326 science facts and 6000 questions, highlighting the gap between AI and human multi-hop reasoning (92%).

Todor Mihaylov, Peter Clark, Tushar Khot et al.

Proposes a denoising NMT training method using trusted data and online data selection, achieving significant BLEU improvements on noisy datasets.

Wei Wang, Taro Watanabe, Macduff Hughes et al.

Using a 2-layer LSTM, the study shows the model detects negative polarity licensing contexts with 89.7% accuracy, confirming syntactic sensitivity.

Jaap Jumelet, Dieuwke Hupkes

ReCode introduces subtree retrieval to enhance neural code generation, achieving up to +2.6 BLEU improvement.

Shirley Anugrah Hayati, Raphael Olivier, Pravalika Avvaru et al.

Proposes a word reward-based length correction to fix length bias and beam search degradation in NMT.

Kenton Murray, David Chiang

Survey on semantic parsing gap in automatic math word problem solvers, highlighting algorithm limitations on large datasets.

Dongxiang Zhang, Lei Wang, Luming Zhang et al.

Adversarial training reveals that protected attributes remain embedded in text representations, making complete removal challenging.

Yanai Elazar, Yoav Goldberg

Using Bayesian Active Learning with Dropout and Bayes-by-Backprop to enhance NLP task performance.

Aditya Siddhant, Zachary C. Lipton

Proposes efficient parallel corpus mining using bilingual sentence embeddings, achieving 48.9% precision (en-fr).

Mandy Guo, Qinlan Shen, Yinfei Yang et al.

Proposes Counterfactual Data Augmentation (CDA) to reduce gender bias in neural NLP, achieving over 80% bias reduction while maintaining accuracy.

Kaiji Lu, Piotr Mardziel, Fangjing Wu et al.

Leveraging WALS and data-driven methods to model language variation, improving multilingual NLP performance.

Edoardo Maria Ponti, Helen O'Horan, Yevgeni Berzak et al.

DecaNLP frames ten NLP tasks as question answering; MQAN uses multi-pointer decoder and coattention, achieving state-of-the-art results in transfer, domain adaptation, and zero-shot learning.

Bryan McCann, Nitish Shirish Keskar, Caiming Xiong et al.

Trained RNN models on CoLA dataset for grammatical acceptability, achieving 67.1% accuracy, below human performance (86.1%).

Alex Warstadt, Amanpreet Singh, Samuel R. Bowman