Papers

Learn more about AI2's Lasting Impact Award
All Projects
All Years
Viewing 21-30 of 461 papers
  • Belief Propagation Neural Networks

    J. Kuck, Shuvam Chakraborty, Hao Tang, R. Luo, Jiaming Song, A. Sabharwal, S. ErmonNeurIPS2020
    Learned neural solvers have successfully been used to solve combinatorial optimization and decision problems. More general counting variants of these problems, however, are still largely solved with hand-crafted solvers. To bridge this gap, we introduce belief propagation neural networks (BPNNs), a class of parameterized operators that operate on factor graphs and generalize Belief Propagation (BP). In its strictest form, a BPNN layer (BPNN-D) is a learned iterative operator that provably maintains many of the desirable properties of BP for any choice of the parameters. Empirically, we show that by training BPNN-D learns to perform the task better than the original BP: it converges 1.7x faster on Ising models while providing tighter bounds. On challenging model counting problems, BPNNs compute estimates 100’s of times faster than state-of-the-art handcrafted methods, while returning an estimate of comparable quality.
  • Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge

    Alon Talmor, Oyvind Tafjord, Peter Clark, Yoav Goldberg, Jonathan BerantNeurIPS • Spotlight Presentation2020
    To what extent can a neural network systematically reason over symbolic facts? Evidence suggests that large pre-trained language models (LMs) acquire some reasoning capacity, but this ability is difficult to control. Recently, it has been shown that Transformer-based models succeed in consistent reasoning over explicit symbolic facts, under a “closed-world" assumption. However, in an open-domain setup, it is desirable to tap into the vast reservoir of implicit knowledge already encoded in the parameters of pre-trained LMs. In this work, we provide a first demonstration that LMs can be trained to reliably perform systematic reasoning combining both implicit, pre-trained knowledge and explicit natural language statements. To do this, we describe a procedure for automatically generating datasets that teach a model new reasoning skills, and demonstrate that models learn to effectively perform inference which involves implicit taxonomic and world knowledge, chaining and counting. Finally, we show that “teaching” the models to reason generalizes beyond the training distribution: they successfully compose the usage of multiple reasoning skills in single examples. Our work paves a path towards open-domain systems that constantly improve by interacting with users who can instantly correct a model by adding simple natural language statements.
  • Do Neural Language Models Overcome Reporting Bias?

    Vered Shwartz and Yejin ChoiProceedings of the 28th International Conference on Computational Linguistics2020
    Mining commonsense knowledge from corpora suffers from reporting bias, over-representing the rare at the expense of the trivial (Gordon and Van Durme, 2013). We study to what extent pre-trained language models overcome this issue. We find that while their generalization capacity allows them to better estimate the plausibility of frequent but unspoken of actions, outcomes, and properties, they also tend to overestimate that of the very rare, amplifying the bias that already exists in their training corpus.
  • Mitigating Biases in CORD-19 for Analyzing COVID-19 Literature

    Anshul Kanakia, Kuansan Wang, Yuxiao Dong, Boya Xie, Kyle Lo, Zhihong Shen, Lucy Lu Wang, Chiyuan Huang, Darrin Eide, Sebastian Kohlmeier, Chieh-Han WuFrontiers in Research Metrics and Analytics2020
    On the behest of the Office of Science and Technology Policy in the White House, six institutions, including ours, have created an open research dataset called COVID-19 Research Dataset (CORD-19) to facilitate the development of question-answering systems that can assist researchers in finding relevant research on COVID-19. As of May 27, 2020, CORD-19 includes more than 100,000 open access publications from major publishers and PubMed as well as preprint articles deposited into medRxiv, bioRxiv, and arXiv. Recent years, however, have also seen question-answering and other machine learning systems exhibit harmful behaviors to humans due to biases in the training data. It is imperative and only ethical for modern scientists to be vigilant in inspecting and be prepared to mitigate the potential biases when working with any datasets. This article describes a framework to examine biases in scientific document collections like CORD-19 by comparing their properties with those derived from the citation behaviors of the entire scientific community. In total, three expanded sets are created for the analyses: 1) the enclosure set CORD-19E composed of CORD-19 articles and their references and citations, mirroring the methodology used in the renowned “A Century of Physics” analysis; 2) the full closure graph CORD-19C that recursively includes references starting with CORD-19; and 3) the inflection closure CORD-19I, that is, a much smaller subset of CORD-19C but already appropriate for statistical analysis based on the theory of the scale-free nature of the citation network. Taken together, all these expanded datasets show much smoother trends when used to analyze global COVID-19 research. The results suggest that while CORD-19 exhibits a strong tilt toward recent and topically focused articles, the knowledge being explored to attack the pandemic encompasses a much longer time span and is very interdisciplinary. A question-answering system with such expanded scope of knowledge may perform better in understanding the literature and answering related questions. However, while CORD-19 appears to have topical coverage biases compared to the expanded sets, the collaboration patterns, especially in terms of team sizes and geographical distributions, are captured very well already in CORD-19 as the raw statistics and trends agree with those from larger datasets.
  • It's not Greek to mBERT: Inducing Word-Level Translations from Multilingual BERT

    Hila Gonen, Shauli Ravfogel, Yanai Elazar, Yoav GoldbergEMNLP • BlackboxNLP Workshop 2020
    Recent works have demonstrated that multilingual BERT (mBERT) learns rich cross-lingual representations, that allow for transfer across languages. We study the word-level translation information embedded in mBERT and present two simple methods that expose remarkable translation capabilities with no fine-tuning. The results suggest that most of this information is encoded in a non-linear way, while some of it can also be recovered with purely linear tools. As part of our analysis, we test the hypothesis that mBERT learns representations which contain both a language-encoding component and an abstract, cross-lingual component, and explicitly identify an empirical language-identity subspace within mBERT representations.
  • Unsupervised Distillation of Syntactic Information from Contextualized Word Representations

    Shauli Ravfogel, Yanai Elazar, Jacob Goldberger, Yoav GoldbergEMNLP • BlackboxNLP Workshop2020
    Contextualized word representations, such as ELMo and BERT, were shown to perform well on various semantic and syntactic task. In this work, we tackle the task of unsupervised disentanglement between semantics and structure in neural language representations: we aim to learn a transformation of the contextualized vectors, that discards the lexical semantics, but keeps the structural information. To this end, we automatically generate groups of sentences which are structurally similar but semantically different, and use metric-learning approach to learn a transformation that emphasizes the structural component that is encoded in the vectors. We demonstrate that our transformation clusters vectors in space by structural properties, rather than by lexical semantics. Finally, we demonstrate the utility of our distilled representations by showing that they outperform the original contextualized representations in a few-shot parsing setting.
  • Neural Natural Language Inference Models Partially Embed Theories of Lexical Entailment and Negation

    Atticus Geiger, Kyle Richardson, Christopher PottsEMNLP • BlackboxNLP Workshop 2020
    We address whether neural models for Natural Language Inference (NLI) can learn the compositional interactions between lexical entailment and negation, using four methods: the behavioral evaluation methods of (1) challenge test sets and (2) systematic generalization tasks, and the structural evaluation methods of (3) probes and (4) interventions. To facilitate this holistic evaluation, we present Monotonicity NLI (MoNLI), a new naturalistic dataset focused on lexical entailment and negation. In our behavioral evaluations, we find that models trained on general-purpose NLI datasets fail systematically on MoNLI examples containing negation, but that MoNLI fine-tuning addresses this failure. In our structural evaluations, we look for evidence that our top-performing BERT-based model has learned to implement the monotonicity algorithm behind MoNLI. Probes yield evidence consistent with this conclusion, and our intervention experiments bolster this, showing that the causal dynamics of the model mirror the causal dynamics of this algorithm on subsets of MoNLI. This suggests that the BERT model at least partially embeds a theory of lexical entailment and negation at an algorithmic level.
  • Document-Level Definition Detection in Scholarly Documents: Existing Models, Error Analyses, and Future Directions

    Dongyeop Kang, Andrew Head, Risham Sidhu, Kyle Lo, Daniel S. Weld, Marti A. HearstEMNLP • SDP workshop2020
    The task of definition detection is important for scholarly papers, because papers often make use of technical terminology that may be unfamiliar to readers. Despite prior work on definition detection, current approaches are far from being accurate enough to use in real-world applications. In this paper, we first perform in-depth error analysis of the current best performing definition detection system and discover major causes of errors. Based on this analysis, we develop a new definition detection system, HEDDEx, that utilizes syntactic features, transformer encoders, and heuristic filters, and evaluate it on a standard sentence-level benchmark. Because current benchmarks evaluate randomly sampled sentences, we propose an alternative evaluation that assesses every sentence within a document. This allows for evaluating recall in addition to precision. HEDDEx outperforms the leading system on both the sentence-level and the document-level tasks, by 12.7 F1 points and 14.4 F1 points, respectively. We note that performance on the high-recall document-level task is much lower than in the standard evaluation approach, due to the necessity of incorporation of document structure as features. We discuss remaining challenges in document-level definition detection, ideas for improvements, and potential issues for the development of reading aid applications.
  • PySBD: Pragmatic Sentence Boundary Disambiguation

    Nipun Sadvilkar, M. NeumannEMNLP • NLP-OSS Workshop2020
    In this paper, we present a rule-based sentence boundary disambiguation Python package that works out-of-the-box for 22 languages. We aim to provide a realistic segmenter which can provide logical sentences even when the format and domain of the input text is unknown. In our work, we adapt the Golden Rules Set (a language-specific set of sentence boundary exemplars) originally implemented as a ruby gem - pragmatic_segmenter - which we ported to Python with additional improvements and functionality. PySBD passes 97.92% of the Golden Rule Set exemplars for English, an improvement of 25% over the next best open-source Python tool.
  • The Extraordinary Failure of Complement Coercion Crowdsourcing

    Yanai Elazar, Victoria Basmov, Shauli Ravfogel, Yoav Goldberg, Reut TsarfatyEMNLP • Insights from Negative Results in NLP Workshop 2020
    Crowdsourcing has eased and scaled up the collection of linguistic annotation in recent years. In this work, we follow known methodologies of collecting labeled data for the complement coercion phenomenon. These are constructions with an implied action -- e.g., "I started a new book I bought last week", where the implied action is reading. We aim to collect annotated data for this phenomenon by reducing it to either of two known tasks: Explicit Completion and Natural Language Inference. However, in both cases, crowdsourcing resulted in low agreement scores, even though we followed the same methodologies as in previous work. Why does the same process fail to yield high agreement scores? We specify our modeling schemes, highlight the differences with previous work and provide some insights about the task and possible explanations for the failure. We conclude that specific phenomena require tailored solutions, not only in specialized algorithms, but also in data collection methods.
All Projects
All Years