Towards Linguistically Generalizable NLP Systems: A Workshop and Shared Task

Ettinger, Allyson; Rao, Sudha; Daumé, Hal; Bender, Emily M.

doi:10.18653/v1/w17-5401

Cited by 74 publications

(66 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As mentioned in the introduction, our approach takes inspiration from "Build it Break it" approaches which have been successfully tried in other domains (Ruef et al, 2016;Ettinger et al, 2017). Those approaches advocate finding faults in systems by having humans look for insecurities (in software) or prediction failures (in models), but do not advocate an automated approach as we do here.…”

Section: Related Workmentioning

confidence: 99%

Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attack

Dinan¹,

Humeau²,

Chintagunta³

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

125

144

View full text Add to dashboard Cite

The detection of offensive language in the context of a dialogue has become an increasingly important application of natural language processing. The detection of trolls in public forums (Galán-García et al., 2016), and the deployment of chatbots in the public domain (Wolf et al., 2017) are two examples that show the necessity of guarding against adversarially offensive behavior on the part of humans. In this work, we develop a training scheme for a model to become robust to such human attacks by an iterative build it, break it, fix it strategy with humans and models in the loop. In detailed experiments we show this approach is considerably more robust than previous systems. Further, we show that offensive language used within a conversation critically depends on the dialogue context, and cannot be viewed as a single sentence offensive detection task as in most previous work. Our newly collected tasks and methods will be made open source and publicly available.

show abstract

Section: Related Workmentioning

confidence: 99%

Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attack

Dinan¹,

Humeau²,

Chintagunta³

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

125

144

View full text Add to dashboard Cite

show abstract

“…In each pair, both annotators annotate the same document, sentence, or token depending on the task. 6 The annotation start by labelling articles in a sample of news articles as containing a protest or not. Sentences of these positively labelled documents are then labelled as containing protest information or not.…”

Section: Datamentioning

confidence: 99%

“…The methods that rely on manual and semi-automated coding, though reliable, require a tremendous amount of effort to replicate on new data as they depend intensely on high quality human effort. On the other hand, text classification and information extraction systems that rely on automated methods yield less reliable results as they tend to perform poorly on texts different from the ones they were developed and validated on [6,12]. The huge amount of news articles that are required to be analyzed and the constant need of repeating the same analyses on new data force us to push limits of automated protest information collection yet again.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Task Set Proposal for Automatic Protest Information Collection Across Multiple Countries

Hürriyetoğlu

Yörük

Yüret

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

We propose a coherent set of tasks for protest information collection in the context of generalizable natural language processing. The tasks are news article classification, event sentence detection, and event extraction. Having tools for collecting event information from data produced in multiple countries enables comparative sociology and politics studies. We have annotated news articles in English from a source and a target country in order to be able to measure the performance of the tools developed using data from one country on data from a different country. Our preliminary experiments have shown that the performance of the tools developed using English texts from India drops to a level that are not usable when they are applied on English texts from China. We think our setting addresses the challenge of building generalizable NLP tools that perform well independent of the source of the text and will accelerate progress in line of developing generalizable NLP systems.

show abstract

“…Seq2Seq based neural architectures have become the go-to architecture to apply to sequence to sequence language tasks. Despite their excellent performance on these tasks, recent work has noted that these models usually do not fully capture the linguistic structure required to generalize beyond the dense sections of the data distribution (Ettinger et al, 2017), and as such, are likely to fail on samples from the tail end of the distribution (such as inputs that are noisy (Belinkov and Bisk, 2018) or of different lengths (Bentivogli et al, 2016)). In this paper, we look at a model's ability to generalize on a simple symbol rewriting task with a clearly defined structure.…”

Section: Introductionmentioning

confidence: 99%

Proceedings of the Workshop on Generalization in the Age of Deep Learning

2018

View full text Add to dashboard Cite

ii Deep learning has brought a wealth of state-of-the-art results and new capabilities. Although methods have achieved near human-level performance on many benchmarks, numerous recent studies imply that these benchmarks only weakly test their intended purpose, and that simple examples produced either by human or machine, cause systems to fail spectacularly. For example, a recently released textual entailment demo was criticized on social media for predicting that "John killed Mary" entails "Mary killed John" with 92% confidence. Such surprising failures combined with the inability to interpret stateof-the-art models have eroded confidence in our systems, and while these systems are not perfect, the real flaw lies with our benchmarks that do not adequately measure a model's ability to generalize, and are thus easily gameable.This workshop provides a venue for exploring new approaches for measuring and enforcing generalization in models. We have solicited work in the following areas:1. Analysis of existing models and their failings.2. Creation of new evaluation paradigms, e.g. zero-shot learning, Winnograd schema, and datasets that avoid explicit types of gamification.3. Modeling advances such as regularization, compositionality, interpretability, inductive bias, multitask learning, and other methods that promote generalization.Our goals are similar in spirit to those of the recent "Build it Break it" shared task. However, we propose going beyond identifying areas of weakness (i.e. "breaking" existing systems), and discussing evaluations that rigorously test generalization as well as modeling techniques for enforcing it.We received eight archival submissions and seven cross submission, accepting five archival papers and all cross submission. Predominately papers covered the first two stated goals of workshop, with the majority identifying flaws in either methods or data. Of the papers proposing new evaluations, many explored using synthetic data. The papers will be presented as posters at the workshop and we are excited to see what discussions they generate. In addition to twelve papers that will be presented we are equally excited for talks from . Finally , we would also like to thank Yejin, Devi and Dan for helping through service on the steering committee.-Yonatan, Omer, Markiii Organizers: AbstractIn this paper, we investigate the tendency of end-to-end neural Machine Reading Comprehension (MRC) models to match shallow patterns rather than perform inference-oriented reasoning on RC benchmarks. We aim to test the ability of these systems to answer questions which focus on referential inference. We propose ParallelQA, a strategy to formulate such questions using parallel passages. We also demonstrate that existing neural models fail to generalize well to this setting. AbstractCommonsense knowledge bases such as Con-ceptNet represent knowledge in the form of relational triples. Inspired by recent work by (Li et al., 2016), we analyse if knowledge base completion models can be used to mine commonsense knowledge fro...

show abstract

Towards Linguistically Generalizable NLP Systems: A Workshop and Shared Task

Cited by 74 publications

References 17 publications

Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attack

Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attack

A Task Set Proposal for Automatic Protest Information Collection Across Multiple Countries

Proceedings of the Workshop on Generalization in the Age of Deep Learning

Contact Info

Product

Resources

About