By allowing models to predict without task-specific training, in-context learning (ICL) with pretrained LLMs has enormous potential in NLP. However, a number of problems persist in ICL. In particular, its performance is sensitive to the choice and order of in-context examples. Given the same set of in-context examples with different orderings, model performance may vary from near random to near state-of-the-art. In this work, we formulate in-context example ordering as an optimization problem. We examine three problem settings that differ in the assumptions they make about what is known about the task. Inspired by the idea of learning from label proportions, we propose two principles for in-context example ordering guided by model’s probability predictions. We apply our proposed principles to thirteen text classification datasets and nine different autoregressive LLMs with 700M to 13B parameters. We demonstrate that our approach outperforms the baselines by improving the classification accuracy, reducing model miscalibration, and also by selecting better in-context examples.

The inevitable appearance of spurious correlations in training datasets hurts the generalization of NLP models on unseen data. Previous work has found that datasets with paired inputs are prone to correlations between a specific part of the input (e.g., the hypothesis in NLI) and the label; consequently, models trained only on those outperform chance. Are these correlations picked up by models trained on the full input data? To address this question, we propose a new evaluation method, Counterfactual Attentiveness Test (CAT). CAT uses counterfactuals by replacing part of the input with its counterpart from a different example (subject to some restrictions), expecting an attentive model to change its prediction. Using CAT, we systematically investigate established supervised and in-context learning models on ten datasets spanning four tasks: natural language inference, reading comprehension, paraphrase detection, and visual & language reasoning. CAT reveals that reliance on such correlations is mainly data-dependent. Surprisingly, we find that GPT3 becomes less attentive with an increased number of demonstrations, while its accuracy on the test data improves. Our results demonstrate that augmenting training or demonstration data with counterfactuals is effective in improving models’ attentiveness. We show that models’ attentiveness measured by CAT reveals different conclusions from solely measuring correlations in data.

Increasingly, model compression techniques enable large language models (LLMs) to be deployed in real-world applications. As a result of this momentum towards local deployment, compressed LLMs will interact with a large population. Prior work on compression typically prioritize preserving perplexity, which is directly analogous to training loss. The impact of compression method on other critical aspects of model behavior—particularly safety—requires systematic assessment. To this end, we investigate the impact of model compression along four dimensions: (1) degeneration harm, i.e., bias and toxicity in generation; (2) representational harm, i.e., biases in discriminative tasks; (3) dialect bias; and (4) language modeling and downstream task performance. We examine a wide spectrum of LLM compression techniques, including unstructured pruning, semi-structured pruning, and quantization. Our analysis reveals that compression can lead to unexpected consequences. Although compression may unintentionally alleviate LLMs’ degeneration harm, it can still exacerbate representational harm. Furthermore, increasing compression produces a divergent impact on different protected groups. Finally, different compression methods have drastically different safety impacts: for example, quantization mostly preserves bias while pruning degrades quickly. Our findings underscore the importance of integrating safety assessments into the development of compressed LLMs to ensure their reliability across real-world applications.

Prompt-based methods have been used extensively across NLP to build zero- and few-shot label predictors. Many NLP tasks are naturally structured: that is, their outputs consist of multiple labels which constrain each other. Annotating data for such tasks can be cumbersome. Can the promise of the prompt-based paradigm be extended to such structured outputs? In this paper, we present a framework for constructing zero- and few-shot linguistic structure predictors. Our key insight is that we can use structural constraints—and combinatorial inference derived from them—to filter out inconsistent structures predicted by large language models. We instantiated this framework on two structured prediction tasks, and five datasets. Across all cases, our results show that enforcing consistency not only constructs structurally valid outputs, but also improves performance over the unconstrained variants.

*Do larger and more performant models resolve NLP’s longstanding robustness issues?* We investigate this question using over 20 models of different sizes spanning different architectural choices and pretraining objectives. We conduct evaluations using (a) out-of-domain and challenge test sets, (b) behavioral testing with CheckLists, (c) contrast sets, and (d) adversarial inputs. Our analysis reveals that not all out-of-domain tests provide insight into robustness. Evaluating with CheckLists and contrast sets shows significant gaps in model performance; merely scaling models does not make them adequately robust. Finally, we point out that current approaches for adversarial evaluations of models are themselves problematic: they can be easily thwarted, and in their current forms, do not represent a sufficiently deep probe of model robustness. We conclude that not only is the question of robustness in NLP as yet unresolved, but even some of the approaches to measure robustness need to be reassessed.

In question answering requiring common sense, language models (e.g., GPT-3) have been used to generate text expressing background knowledge that helps improve performance. Yet the cost of working with such models is very high; in this work, we finetune smaller language models to generate useful intermediate context, referred to here as elaborations. Our framework alternates between updating two language models—an elaboration generator and an answer predictor—allowing each to influence the other. Using less than 0.5% of the parameters of GPT-3, our model outperforms alternatives with similar sizes and closes the gap with GPT-3 on four commonsense question answering benchmarks. Human evaluations show that the quality of the generated elaborations is high.

Context is everything, even in commonsense moral reasoning. Changing contexts can flip the moral judgment of an action; Lying to a friend is wrong in general, but may be morally acceptable if it is intended to protect their life. We present ClarifyDelphi, an interactive system that learns to ask clarification questions (e.g., why did you lie to your friend?) in order to elicit additional salient contexts of a social or moral situation. We posit that questions whose potential answers lead to diverging moral judgments are the most informative. Thus, we propose a reinforcement learning framework with a defeasibility reward that aims to maximize the divergence between moral judgments of hypothetical answers to a question. Human evaluation demonstrates that our system generates more relevant, informative and defeasible questions compared to competitive baselines. Our work is ultimately inspired by studies in cognitive science that have investigated the flexibility in moral cognition (i.e., the diverse contexts in which moral rules can be bent), and we hope that research in this direction can assist both cognitive and computational investigations of moral judgments.

Ensuring the effectiveness of text-based crisis counseling requires observing ongoing conversations and providing feedback, both labor-intensive tasks. Automatic analysis of conversations—at the full chat and utterance levels—may help support counselors and provide better care. While some session-level training data (e.g., rating of patient risk) is often available from counselors, labeling utterances requires expensive post hoc annotation. But the latter can not only provide insights about conversation dynamics, but can also serve to support quality assurance efforts for counselors. In this paper, we examine if inexpensive—and potentially noisy—session-level annotation can help improve label utterances. To this end, we propose a logic-based indirect supervision approach that exploits declaratively stated structural dependencies between both levels of annotation to improve utterance modeling. We show that adding these rules gives an improvement of 3.5% f-score over a strong multi-task baseline for utterance-level predictions. We demonstrate via ablation studies how indirect supervision via logic rules also improves the consistency and robustness of the system.

Can language models transform inputs to protect text classifiers against adversarial attacks? In this work, we present ATINTER, a model that intercepts and learns to rewrite adversarial inputs to make them non-adversarial for a downstream text classifier. Our experiments on four datasets and five attack mechanisms reveal that ATINTER is effective at providing better adversarial robustness than existing defense approaches, without compromising task accuracy. For example, on sentiment classification using the SST-2 dataset, our method improves the adversarial accuracy over the best existing defense approach by more than 4% with a smaller decrease in task accuracy (0.5 % vs 2.5%). Moreover, we show that ATINTER generalizes across multiple downstream tasks and classifiers without having to explicitly retrain it for those settings. For example, we find that when ATINTER is trained to remove adversarial perturbations for the sentiment classification task on the SST-2 dataset, it even transfers to a semantically different task of news classification (on AGNews) and improves the adversarial robustness by more than 10%.

Good datasets are a foundation of NLP research, and form the basis for training and evaluating models of language use. While creating datasets, the standard practice is to verify the annotation consistency using a committee of human annotators. This norm assumes that multiple annotators are available, which is not the case for highly specialized tasks or low-resource languages. In this paper, we ask: Can we evaluate the quality of a dataset constructed by a single human annotator? To address this question, we propose four weak verifiers to help estimate dataset quality, and outline when each may be employed. We instantiate these strategies for the task of semantic analysis of adpositions in Gujarati, a low-resource language, and show that our weak verifiers concur with a double-annotation study. As an added contribution, we also release the first dataset with semantic annotations in Gujarati along with several model baselines.

Semantic role labeling (SRL) has multiple disjoint label sets, e.g., VerbNet and PropBank. Creating these datasets is challenging, therefore a natural question is how to use each one to help the other. Prior work has shown that cross-task interaction helps, but only explored multitask learning so far. A common issue with multi-task setup is that argument sequences are still separately decoded, running the risk of generating structurally inconsistent label sequences (as per lexicons like Semlink). In this paper, we eliminate such issue with a framework that jointly models VerbNet and PropBank labels as one sequence. In this setup, we show that enforcing Semlink constraints during decoding constantly improves the overall F1. With special input constructions, our joint model infers VerbNet arguments from given PropBank arguments with over 99 F1. For learning, we propose a constrained marginal model that learns with knowledge defined in Semlink to further benefit from the large amounts of PropBank-only data. On the joint benchmark based on CoNLL05, our models achieve state-of-the-art F1’s, outperforming the prior best in-domain model by 3.5 (VerbNet) and 0.8 (PropBank). For out-of-domain generalization, our models surpass the prior best by 3.4 (VerbNet) and 0.2 (PropBank).

Semi-structured data, such as Infobox tables, often include temporal information about entities, either implicitly or explicitly. Can current NLP systems reason about such information in semi-structured tables? To tackle this question, we introduce the task of temporal question answering on semi-structured tables. We present a dataset, TEMPTABQA, which comprises 11,454 question-answer pairs extracted from 1,208 Wikipedia Infobox tables spanning more than 90 distinct domains. Using this dataset, we evaluate several state-of-the-art models for temporal reasoning. We observe that even the top-performing LLMs lag behind human performance by more than 13.5 F1 points. Given these results, our dataset has the potential to serve as a challenging benchmark to improve the temporal reasoning capabilities of NLP models.

Probing contextualized representations typically involves comparing task-specific model predictions against ground truth linguistic labels. Although this methodology shows what information can be recovered by a classifier, it does not reveal how a classifier uses the representation to make its decision. To address the latter problem, we ask: Do task-classifiers rely on representation- and task-independent geometric patterns in the embedding space? We explore this question by developing MetaProbe, an approach that uses geometric properties of representations to predict the behavior of task-specific classifiers (i.e., their predictions as opposed to the ground truth). Our experiments reveal the existence of universal geometric patterns across representations that can predict classifier predictions. Consequently, this allows us to posit a geometric explanation for the impressive performance of contextualized representations.

There are many different forms of psychotherapy. Itemized inventories of psychotherapeutic interventions provide a mechanism for evaluating the quality of care received by clients and for conducting research on how psychotherapy helps. However, evaluations such as these are slow, expensive, and are rarely used outside of well-funded research studies. Natural language processing research has progressed to allow automating such tasks. Yet, NLP work in this area has been restricted to evaluating a single approach to treatment, when prior research indicates therapists used a wide variety of interventions with their clients, often in the same session. In this paper, we frame this scenario as a multi-label classification task, and develop a group of models aimed at predicting a wide variety of therapist talk-turn level orientations. Our models achieve F1 macro scores of 0.5, with the class F1 ranging from 0.36 to 0.67. We present analyses which offer insights into the capability of such models to capture psychotherapy approaches, and which may complement human judgment.

Given the prevalence of pre-trained contextualized representations in today’s NLP, there have been many efforts to understand what information they contain, and why they seem to be universally successful. The most common approach to use these representations involves fine-tuning them for an end task. Yet, how fine-tuning changes the underlying embedding space is less studied. In this work, we study the English BERT family and use two probing techniques to analyze how fine-tuning changes the space. We hypothesize that fine-tuning affects classification performance by increasing the distances between examples associated with different labels. We confirm this hypothesis with carefully designed experiments on five different NLP tasks. Via these experiments, we also discover an exception to the prevailing wisdom that “fine-tuning always improves performance”. Finally, by comparing the representations before and after fine-tuning, we discover that fine-tuning does not introduce arbitrary changes to representations; instead, it adjusts the representations to downstream tasks while largely preserving the original spatial structure of the data points.

When pre-trained contextualized embedding-based models developed for unstructured data are adapted for structured tabular data, they perform admirably. However, recent probing studies show that these models use spurious correlations, and often predict inference labels by focusing on false evidence or ignoring it altogether. To study this issue, we introduce the task of Trustworthy Tabular Reasoning, where a model needs to extract evidence to be used for reasoning, in addition to predicting the label. As a case study, we propose a two-stage sequential prediction approach, which includes an evidence extraction and an inference stage. First, we crowdsource evidence row labels and develop several unsupervised and supervised evidence extraction strategies for InfoTabS, a tabular NLI benchmark. Our evidence extraction strategy outperforms earlier baselines. On the downstream tabular inference task, using only the automatically extracted evidence as the premise, our approach outperforms prior benchmarks.

The SNACS framework provides a network of semantic labels called supersenses for annotating adpositional semantics in corpora. In this work, we consider English prepositions (and prepositional phrases) that are chiefly pragmatic, contributing extra-propositional contextual information such as speaker attitudes and discourse structure. We introduce a preliminary taxonomy of pragmatic meanings to supplement the semantic SNACS supersenses, with guidelines for the annotation of coherence connectives, commentary markers, and topic and focus markers. We also examine annotation disagreements, delve into the trickiest boundary cases, and offer a discussion of future improvements.

Neural models command state-of-the-art performance across NLP tasks, including ones involving “reasoning”. Models claiming to reason about the evidence presented to them should attend to the correct parts of the input while avoiding spurious patterns therein, be self-consistent in their predictions across inputs, and be immune to biases derived from their pre-training in a nuanced, context- sensitive fashion. Do the prevalent *BERT- family of models do so? In this paper, we study this question using the problem of reasoning on tabular data. Tabular inputs are especially well-suited for the study—they admit systematic probes targeting the properties listed above. Our experiments demonstrate that a RoBERTa-based model, representative of the current state-of-the-art, fails at reasoning on the following counts: it (a) ignores relevant parts of the evidence, (b) is over- sensitive to annotation artifacts, and (c) relies on the knowledge encoded in the pre-trained language model rather than the evidence presented in its tabular inputs. Finally, through inoculation experiments, we show that fine- tuning the model on perturbed data does not help it overcome the above challenges.

In this work, we introduce : the largest publicly available multilingual dataset for factual verification of naturally existing real-world claims. The dataset contains short statements in 25 languages and is labeled for veracity by expert fact-checkers. The dataset includes a multilingual evaluation benchmark that measures both out-of-domain generalization, and zero-shot capabilities of the multilingual models. Using state-of-the-art multilingual transformer-based models, we develop several automated fact-checking models that, along with textual claims, make use of additional metadata and evidence from news stories retrieved using a search engine. Empirically, our best model attains an F-score of around 40%, suggesting that our dataset is a challenging benchmark for the evaluation of multilingual fact-checking models.

Although current CCG supertaggers achieve high accuracy on the standard WSJ test set, few systems make use of the categories’ internal structure that will drive the syntactic derivation during parsing. The tagset is traditionally truncated, discarding the many rare and complex category types in the long tail. However, supertags are themselves trees. Rather than give up on rare tags, we investigate constructive models that account for their internal structure, including novel methods for tree-structured prediction. Our best tagger is capable of recovering a sizeable fraction of the long-tail supertags and even generates CCG categories that have never been seen in training, while approximating the prior state of the art in overall tag accuracy with fewer parameters. We further investigate how well different approaches generalize to out-of-domain evaluation sets.

Reasoning about tabular information presents unique challenges to modern NLP approaches which largely rely on pre-trained contextualized embeddings of text. In this paper, we study these challenges through the problem of tabular natural language inference. We propose easy and effective modifications to how information is presented to a model for this task. We show via systematic experiments that these strategies substantially improve tabular inference performance.

Understanding how linguistic structure is encoded in contextualized embedding could help explain their impressive performance across NLP. Existing approaches for probing them usually call for training classifiers and use the accuracy, mutual information, or complexity as a proxy for the representation’s goodness. In this work, we argue that doing so can be unreliable because different representations may need different classifiers. We develop a heuristic, DirectProbe, that directly studies the geometry of a representation by building upon the notion of a version space for a task. Experiments with several linguistic tasks and contextualized embeddings show that, even without training classifiers, DirectProbe can shine lights on how an embedding space represents labels and also anticipate the classifier performance for the representation.

Tracking entity states is a natural language processing task assumed to require human annotation. In order to reduce the time and expenses associated with annotation, we introduce a new method to automatically extract entity states, including location and existence state of entities, following Dalvi et al. (2018) and Tandon et al. (2020). For this purpose, we rely primarily on the semantic representations generated by the state of the art VerbNet parser (Gung, 2020), and extract the entities (event participants) and their states, based on the semantic predicates of the generated VerbNet semantic representation, which is in propositional logic format. For evaluation, we used ProPara (Dalvi et al., 2018), a reading comprehension dataset which is annotated with entity states in each sentence, and tracks those states in paragraphs of natural human-authored procedural texts. Given the presented limitations of the method, the peculiarities of the ProPara dataset annotations, and that our system, Lexis, makes no use of task-specific training data and relies solely on VerbNet, the results are promising, showcasing the value of lexical resources.

Language representations are known to carry stereotypical biases and, as a result, lead to biased predictions in downstream tasks. While existing methods are effective at mitigating biases by linear projection, such methods are too aggressive: they not only remove bias, but also erase valuable information from word embeddings. We develop new measures for evaluating specific information retention that demonstrate the tradeoff between bias removal and information retention. To address this challenge, we propose OSCaR (Orthogonal Subspace Correction and Rectification), a bias-mitigating method that focuses on disentangling biased associations between concepts instead of removing concepts wholesale. Our experiments on gender biases show that OSCaR is a well-balanced approach that ensures that semantic information is retained in the embeddings and bias is also effectively mitigated.

We present a method for exploring regions around individual points in a contextualized vector space (particularly, BERT space), as a way to investigate how these regions correspond to word senses. By inducing a contextualized “pseudoword” vector as a stand-in for a static embedding in the input layer, and then performing masked prediction of a word in the sentence, we are able to investigate the geometry of the BERT-space in a controlled manner around individual instances. Using our method on a set of carefully constructed sentences targeting highly ambiguous English words, we find substantial regularity in the contextualized space, with regions that correspond to distinct word senses; but between these regions there are occasionally “sense voids”—regions that do not correspond to any intelligible sense.

In this paper, we observe that semi-structured tabulated text is ubiquitous; understanding them requires not only comprehending the meaning of text fragments, but also implicit relationships between them. We argue that such data can prove as a testing ground for understanding how we reason about information. To study this, we introduce a new dataset called INFOTABS, comprising of human-written textual hypotheses based on premises that are tables extracted from Wikipedia info-boxes. Our analysis shows that the semi-structured, multi-domain and heterogeneous nature of the premises admits complex, multi-faceted reasoning. Experiments reveal that, while human annotators agree on the relationships between a table-hypothesis pair, several standard modeling strategies are unsuccessful at the task, suggesting that reasoning about tables can pose a difficult modeling challenge.

Various natural language processing tasks are structured prediction problems where outputs are constructed with multiple interdependent decisions. Past work has shown that domain knowledge, framed as constraints over the output space, can help improve predictive accuracy. However, designing good constraints often relies on domain expertise. In this paper, we study the problem of learning such constraints. We frame the problem as that of training a two-layer rectifier network to identify valid structures or substructures, and show a construction for converting a trained network into a system of linear constraints over the inference variables. Our experiments on several NLP tasks show that the learned constraints can improve the prediction accuracy, especially when the number of training examples is small.

Recent neural network-driven semantic role labeling (SRL) systems have shown impressive improvements in F1 scores. These improvements are due to expressive input representations, which, at least at the surface, are orthogonal to knowledge-rich constrained decoding mechanisms that helped linear SRL models. Introducing the benefits of structure to inform neural models presents a methodological challenge. In this paper, we present a structured tuning framework to improve models using softened constraints only at training time. Our framework leverages the expressiveness of neural networks and provides supervision with structured loss components. We start with a strong baseline (RoBERTa) to validate the impact of our approach, and show that our framework outperforms the baseline by learning to comply with declarative constraints. Additionally, our experiments with smaller training sizes show that we can achieve consistent improvements under low-resource scenarios.

We reevaluate an existing adpositional annotation scheme with respect to two thorny semantic domains: accompaniment and purpose. ‘Accompaniment’ broadly speaking includes two entities situated together or participating in the same event, while ‘purpose’ broadly speaking covers the desired outcome of an action, the intended use or evaluated use of an entity, and more. We argue the policy in the SNACS scheme for English should be recalibrated with respect to these clusters of interrelated meanings without adding complexity to the overall scheme. Our analysis highlights tradeoffs in lumping vs. splitting decisions as well as the flexibility afforded by the construal analysis.

While language embeddings have been shown to have stereotyping biases, how these biases affect downstream question answering (QA) models remains unexplored. We present UNQOVER, a general framework to probe and quantify biases through underspecified questions. We show that a naive use of model scores can lead to incorrect bias estimates due to two forms of reasoning errors: positional dependence and question independence. We design a formalism that isolates the aforementioned errors. As case studies, we use this metric to analyze four important classes of stereotypes: gender, nationality, ethnicity, and religion. We probe five transformer-based QA models trained on two QA datasets, along with their underlying language models. Our broad study reveals that (1) all these models, with and without fine-tuning, have notable stereotyping biases in these classes; (2) larger models often have higher bias; and (3) the effect of fine-tuning on bias varies strongly with the dataset and the model size.

Most word embeddings today are trained by optimizing a language modeling goal of scoring words in their context, modeled as a multi-class classification problem. In this paper, we argue that, despite the successes of this assumption, it is incomplete: in addition to its context, orthographical or morphological aspects of words can offer clues about their meaning. We define a new modeling framework for training word embeddings that captures this intuition. Our framework is based on the well-studied problem of multi-label classification and, consequently, exposes several design choices for featurizing words and contexts, loss functions for training and score normalization. Indeed, standard models such as CBOW and fasttext are specific choices along each of these axes. We show via experiments that by combining feature engineering with embedding learning, our method can outperform CBOW using only 10% of the training data in both the standard word embedding evaluations and also text classification experiments.

Today, the dominant paradigm for training neural networks involves minimizing task loss on a large dataset. Using world knowledge to inform a model, and yet retain the ability to perform end-to-end training remains an open question. In this paper, we present a novel framework for introducing declarative knowledge to neural network architectures in order to guide training and prediction. Our framework systematically compiles logical statements into computation graphs that augment a neural network without extra learnable parameters or manual redesign. We evaluate our modeling strategy on three tasks: machine comprehension, natural language inference, and text chunking. Our experiments show that knowledge-augmented networks can strongly improve over baselines, especially in low-data regimes.

Automatically analyzing dialogue can help understand and guide behavior in domains such as counseling, where interactions are largely mediated by conversation. In this paper, we study modeling behavioral codes used to asses a psychotherapy treatment style called Motivational Interviewing (MI), which is effective for addressing substance abuse and related problems. Specifically, we address the problem of providing real-time guidance to therapists with a dialogue observer that (1) categorizes therapist and client MI behavioral codes and, (2) forecasts codes for upcoming utterances to help guide the conversation and potentially alert the therapist. For both tasks, we define neural network models that build upon recent successes in dialogue modeling. Our experiments demonstrate that our models can outperform several baselines for both tasks. We also report the results of a careful analysis that reveals the impact of the various network design tradeoffs for modeling therapy dialogue.

One of the goals of natural language understanding is to develop models that map sentences into meaning representations. However, training such models requires expensive annotation of complex structures, which hinders their adoption. Learning to actively-learn(LTAL) is a recent paradigm for reducing the amount of labeled data by learning a policy that selects which samples should be labeled. In this work, we examine LTAL for learning semantic representations, such as QA-SRL. We show that even an oracle policy that is allowed to pick examples that maximize performance on the test set (and constitutes an upper bound on the potential of LTAL), does not substantially improve performance compared to a random policy. We investigate factors that could explain this finding and show that a distinguishing characteristic of successful applications of LTAL is the interaction between optimization and the oracle policy selection process. In successful applications of LTAL, the examples selected by the oracle policy do not substantially depend on the optimization procedure, while in our setup the stochastic nature of optimization strongly affects the examples selected by the oracle. We conclude that the current applicability of LTAL for improving data efficiency in learning semantic meaning representations is limited.

This paper describes the system submission of our team Amazon to the shared task on Cross Framework Meaning Representation Parsing (MRP) at the 2019 Conference for Computational Language Learning (CoNLL). Via extensive analysis of implicit alignments in AMR, we recategorize five meaning representations (MRs) into two classes: Lexical- Anchoring and Phrasal-Anchoring. Then we propose a unified graph-based parsing framework for the lexical-anchoring MRs, and a phrase-structure parsing for one of the phrasal- anchoring MRs, UCCA. Our system submission ranked 1st in the AMR subtask, and later improvements show promising results on other frameworks as well.

While neural models show remarkable accuracy on individual predictions, their internal beliefs can be inconsistent across examples. In this paper, we formalize such inconsistency as a generalization of prediction error. We propose a learning framework for constraining models using logic rules to regularize them away from inconsistency. Our framework can leverage both labeled and unlabeled examples and is directly compatible with off-the-shelf learning schemes without model redesign. We instantiate our framework on natural language inference, where experiments show that enforcing invariants stated in logic can help make the predictions of neural models both accurate and consistent.

Research on adpositions and possessives in multiple languages has led to a small inventory of general-purpose meaning classes that disambiguate tokens. Importantly, that work has argued for a principled separation of the semantic role in a scene from the function coded by morphosyntax. Here, we ask whether this approach can be generalized beyond adpositions and possessives to cover all scene participants—including subjects and objects—directly, without reference to a frame lexicon. We present new guidelines for English and the results of an interannotator agreement study.

Semantic relations are often signaled with prepositional or possessive marking—but extreme polysemy bedevils their analysis and automatic interpretation. We introduce a new annotation scheme, corpus, and task for the disambiguation of prepositions and possessives in English. Unlike previous approaches, our annotations are comprehensive with respect to types and tokens of these markers; use broadly applicable supersense classes rather than fine-grained dictionary definitions; unite prepositions and possessives under the same class inventory; and distinguish between a marker’s lexical contribution and the role it marks in the context of a predicate or scene. Strong interannotator agreement rates, as well as encouraging disambiguation results with established supervised methods, speak to the viability of the scheme and task.

Neural networks models have gained unprecedented popularity in natural language processing due to their state-of-the-art performance and the flexible end-to-end training scheme. Despite their advantages, the lack of interpretability hinders the deployment and refinement of the models. In this work, we present a flexible visualization library for creating customized visual analytic environments, in which the user can investigate and interrogate the relationships among the input, the model internals (i.e., attention), and the output predictions, which in turn shed light on the model decision-making process.

We consider the semantics of prepositions, revisiting a broad-coverage annotation scheme used for annotating all 4,250 preposition tokens in a 55,000 word corpus of English. Attempts to apply the scheme to adpositions and case markers in other languages, as well as some problematic cases in English, have led us to reconsider the assumption that an adposition’s lexical contribution is equivalent to the role/relation that it mediates. Our proposal is to embrace the potential for construal in adposition use, expressing such phenomena directly at the token level to manage complexity and avoid sense proliferation. We suggest a framework to represent both the scene role and the adposition’s lexical function so they can be annotated at scale—supporting automatic, statistical processing of domain-general language—and discuss how this representation would allow for a simpler inventory of labels.

Making decisions in natural language processing problems often involves assigning values to sets of interdependent variables where the expressive dependency structure can influence, or even dictate what assignments are possible. This setting includes a broad range of structured prediction problems such as semantic role labeling, named entity and relation recognition, co-reference resolution, dependency parsing and semantic parsing. The setting is also appropriate for cases that may require making global decisions that involve multiple components, possibly pre-designed or pre-learned, as in event recognition and analysis, summarization, paraphrasing, textual entailment and question answering. In all these cases, it is natural to formulate the decision problem as a constrained optimization problem, with an objective function that is composed of learned models, subject to domain or problem specific constraints.Over the last few years, starting with a couple of papers written by (Roth & Yih, 2004, 2005), dozens of papers have been using the Integer linear programming (ILP) formulation developed there, including several award-winning papers (e.g., (Martins, Smith, & Xing, 2009; Koo, Rush, Collins, Jaakkola, & Sontag., 2010; Berant, Dagan, & Goldberger, 2011)).This tutorial will present the key ingredients of ILP formulations of natural language processing problems, aiming at guiding readers through the key modeling steps, explaining the learning and inference paradigms and exemplifying these by providing examples from the literature. We will cover a range of topics, from the theoretical foundations of learning and inference with ILP models, to practical modeling guides, to software packages and applications.The goal of this tutorial is to introduce the computational framework to broader ACL community, motivate it as a generic framework for learning and inference in global NLP decision problems, present some of the key theoretical and practical issues involved and survey some of the existing applications of it as a way to promote further development of the framework and additional applications. We will also make connections with some of the “hot” topics in current NLP research and show how they can be used within the general framework proposed here. The tutorial will thus be useful for many of the senior and junior researchers that have interest in global decision problems in NLP, providing a concise overview of recent perspectives and research results.

Though feature extraction is a necessary first step in statistical NLP, it is often seen as a mere preprocessing step. Yet, it can dominate computation time, both during training, and especially at deployment. In this paper, we formalize feature extraction from an algebraic perspective. Our formalization allows us to define a message passing algorithm that can restructure feature templates to be more computationally efficient. We show via experiments on text chunking and relation extraction that this restructuring does indeed speed up feature extraction in practice by reducing redundant computation.

When designing Natural Language Processing (NLP) applications that use Machine Learning (ML) techniques, feature extraction becomes a significant part of the development effort, whether developing a new application or attempting to reproduce results reported for existing NLP tasks. We present EDISON, a Java library of feature generation functions used in a suite of state-of-the-art NLP tools, based on a set of generic NLP data structures. These feature extractors populate simple data structures encoding the extracted features, which the package can also serialize to an intuitive JSON file format that can be easily mapped to formats used by ML packages. EDISON can also be used programmatically with JVM-based (Java/Scala) NLP software to provide the feature extractor input. The collection of feature extractors is organised hierarchically and a simple search interface is provided. In this paper we include examples that demonstrate the versatility and ease-of-use of the EDISON feature extraction suite to show that this can significantly reduce the time spent by developers on feature extraction design for NLP systems. The library is publicly hosted at https://github.com/IllinoisCogComp/illinois-cogcomp-nlp/, and we hope that other NLP researchers will contribute to the set of feature extractors. In this way, the community can help simplify reproduction of published results and the integration of ideas from diverse sources when developing new and improved NLP applications.

This paper introduces the problem of predicting semantic relations expressed by prepositions and develops statistical learning models for predicting the relations, their arguments and the semantic types of the arguments. We define an inventory of 32 relations, building on the word sense disambiguation task for prepositions and collapsing related senses across prepositions. Given a preposition in a sentence, our computational task to jointly model the preposition relation and its arguments along with their semantic types, as a way to support the relation prediction. The annotated data, however, only provides labels for the relation label, and not the arguments and types. We address this by presenting two models for preposition relation labeling. Our generalization of latent structure SVM gives close to 90% accuracy on relation labeling. Further, by jointly predicting the relation, arguments, and their types along with preposition sense, we show that we can not only improve the relation accuracy, but also significantly improve sense prediction accuracy.

Natural Language Processing continues to grow in popularity in a range of research and commercial applications, yet managing the wide array of potential NLP components remains a difficult problem. This paper describes Curator, an NLP management framework designed to address some common problems and inefficiencies associated with building NLP process pipelines; and Edison, an NLP data structure library in Java that provides streamlined interactions with Curator and offers a range of useful supporting functionality.