Thailand

Pre-trained language models have achieved human-level performance on many Machine Reading Comprehension (MRC) tasks, but it remains unclear whether these models truly understand language or answer questions by exploiting statistical biases in datasets. Here, we demonstrate a simple yet effective method to attack MRC models and reveal the statistical biases in these models. We apply the method to the RACE dataset, for which the answer to each MRC question is selected from 4 options. It is found that several pre-trained language models, including BERT, ALBERT, and RoBERTa, show consistent preference to some options, even when these options are irrelevant to the question. When interfered by these irrelevant options, the performance of MRC models can be reduced from human-level performance to the chance-level performance. Human readers, however, are not clearly affected by these irrelevant options. Finally, we propose an augmented training method that can greatly reduce models&#39; statistical biases.

ACL 2021

Using Adversarial Attacks to Reveal the Statistical Bias in Machine Reading Comprehension Models

statistical bias

race

machine reading comprehension

adversarial attack

**Welcome to ACL-IJCNLP 2021!**

The great event is jointly organized by the Association for Computational Linguistics (ACL) and Asian Federation of Natural Language Processing (AFNLP). 

This Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (ACL-IJCNLP 2021) was to be held in Bangkok, Thailand, during August 1-6, 2021 and is now virtual on the Underline platform.

As in previous years, the program of the conference includes a poster session, tutorials, workshops and demonstrations in addition to the main conference.


We were able to keep the registration fees similar to those charged for the virtual ACL 2020. The one fee allows attendance at the main conference and any/all tutorials and workshops. These fees would be $125 Regular Early and $175 Regular Late; $50 Student Early and $75 Student Late. Early registration closes at midnight July 11, 2021 (Eastern Daylight Time).

**Reminder:** It is ACL’s policy that at least one author of each accepted paper (including ACL Finding papers) must register for the conference.

**Reminder2:** Underline site will open closer to the event. If you already registered you will receive access detail

Registration is now open

Pre-trained language models have achieved human-level performance on many Machine Reading Comprehension (MRC) tasks, but it remains unclear whether these models truly understand language or answer questions by exploiting statistical biases in datasets. Here, we demonstrate a simple yet effective method to attack MRC models and reveal the statistical biases in these models. We apply the method to the RACE dataset, for which the answer to each MRC question is selected from 4 options. It is found that several pre-trained language models, including BERT, ALBERT, and RoBERTa, show consistent preference to some options, even when these options are irrelevant to the question. When interfered by these irrelevant options, the performance of MRC models can be reduced from human-level performance to the chance-level performance. Human readers, however, are not clearly affected by these irrelevant options. Finally, we propose an augmented training method that can greatly reduce models' statistical biases.

technical paper

Event extraction has long been a challenging task, addressed mostly with supervised methods that require expensive annotation and are not extensible to new event ontologies. In this work, we explore the possibility of zero-shot event extraction by formulating it as a set of Textual Entailment (TE) and/or Question Answering (QA) queries (e.g. "A city was attacked" entails "There is an attack"), exploiting pretrained TE/QA models for direct transfer. On ACE-2005 and ERE, our system achieves acceptable results, yet there is still a large gap from supervised approaches, showing that current QA and TE technologies fail in transferring to a different domain. To investigate the reasons behind the gap, we analyze the remaining key challenges, their respective impact, and possible improvement directions.

Zero-shot Event Extraction via Transfer Learning: Challenges and Insights

Does the effectiveness of neural language models derive entirely from accurate modeling of surface word co-occurrence statistics, or do these models represent and reason about the world they describe? In BART and T5 transformer language models, we identify contextual word representations that function as *models of entities and situations* as they evolve throughout a discourse. These neural representations have functional similarities to linguistic models of dynamic semantics: they support a linear readout of each entity's current properties and relations, and can be manipulated with predictable effects on language generation. Our results indicate that prediction in pretrained neural language models is supported, at least in part, by dynamic representations of meaning and implicit simulation of entity state, and that this behavior can be learned with only text as training data.

Implicit Representations of Meaning in Neural Language Models

Targeted syntactic evaluations have demonstrated the ability of language models to perform subject-verb agreement given difficult contexts. 
To elucidate the mechanisms by which the models accomplish this behavior, this study applies causal mediation analysis to pre-trained neural language models.
We investigate the magnitude of models' preferences for grammatical inflections, as well as whether neurons process subject-verb agreement similarly across sentences with different syntactic structures.
We uncover similarities and differences across architectures and model sizes—notably, that larger models do not necessarily learn stronger preferences.
We also observe two distinct mechanisms for producing subject-verb agreement depending on the syntactic structure of the input sentence.
Finally, we find that language models rely on similar sets of neurons when given sentences with similar syntactic structure.

Causal Analysis of Syntactic Agreement Mechanisms in Neural Language Models

NLP has a rich history of representing our prior understanding of language in the form of graphs. Recent work on analyzing contextualized text representations has focused on hand-designed probe models to understand how and to what extent do these representations encode a particular linguistic phenomenon. However, due to the inter-dependence of various phenomena and randomness of training probe models, detecting how these representations encode the rich information in these linguistic graphs remains a challenging problem. In this paper, we propose a new information-theoretic probe, Bird's Eye, which is a fairly simple probe method for detecting if and how these representations encode the information in these linguistic graphs. Instead of using model performance, our probe takes an information-theoretic view of probing and estimates the mutual information between the linguistic graph embedded in a continuous space and the contextualized word representations. Furthermore, we also propose an approach to use our probe to investigate localized linguistic information in the linguistic graphs using perturbation analysis. We call this probing setup Worm's Eye. Using these probes, we analyze the BERT models on its ability to encode a syntactic and a semantic graph structure, and find that these models encode to some degree both syntactic as well as semantic information; albeit syntactic information to a greater extent.

Bird’s Eye: Probing for Linguistic Graph Structures with a Simple Information-Theoretic Approach

Previous literatures show that pre-trained masked language models (MLMs) such as BERT can achieve competitive factual knowledge extraction performance on some datasets, indicating that MLMs can potentially be a reliable knowledge source. In this paper, we conduct a rigorous study to explore the underlying predicting mechanisms of MLMs over different extraction paradigms. By investigating the behaviors of MLMs, we find that previous decent performance mainly owes to the biased prompts which overfit dataset artifacts. Furthermore, incorporating illustrative cases and external contexts improve knowledge prediction mainly due to entity type guidance and golden answer leakage. Our findings shed light on the underlying predicting mechanisms of MLMs, and strongly question the previous conclusion that current MLMs can potentially serve as reliable factual knowledge bases.

Knowledgeable or Educated Guess? Revisiting Language Models as Knowledge Bases

We study the problem of generating data poisoning attacks against Knowledge Graph Embedding (KGE) models for the task of link prediction in knowledge graphs. To poison KGE models, we propose to exploit their inductive abilities which are captured through the relationship patterns like symmetry, inversion and composition in the knowledge graph. Specifically, to degrade the model's prediction confidence on target facts, we propose to improve the model's prediction confidence on a set of decoy facts. Thus, we craft adversarial additions that can improve the model's prediction confidence on decoy facts through different inference patterns. Our experiments demonstrate that the proposed poisoning attacks outperform state-of-art baselines on four KGE models for two publicly available datasets. We also find that the symmetry pattern based attacks generalize across all model-dataset combinations which indicates the sensitivity of KGE models to this pattern.

Poisoning Knowledge Graph Embeddings via Relation Inference Patterns

Target meaning representations for semantic parsing tasks are often based on programming or query languages, such as SQL, and can be formalized by a context-free grammar. Assuming a priori knowledge of the target domain, such grammars can be exploited to enforce syntactical constraints when predicting logical forms. To that end, we assess how syntactical parsers can be integrated into modern encoder-decoder frameworks. Specifically, we implement an attentional Seq2Seq model that uses an LR parser to maintain syntactically valid sequences throughout the decoding procedure. Compared to other approaches to grammar-guided decoding that modify the underlying neural network architecture or attempt to derive full parse trees, our approach is conceptually simpler, adds less computational overhead during inference and integrates seamlessly with current Seq2Seq frameworks. We present preliminary evaluation results against a recurrent Seq2Seq baseline on GeoQuery and ATIS and demonstrate improved performance while enforcing grammatical constraints.

Grammar-Constrained Neural Semantic Parsing with LR Parsers

In this paper, we introduce the new task of controllable text edition, in which we take as input a long text, a question, and a target answer, and the output is a minimally modified text, so that it fits the target answer. This task is very important in many situations, such as changing some conditions, consequences, or properties in a legal document, or changing some key information of an event in a news text. This is very challenging, as it is hard to obtain a parallel corpus for training, and we need to first find all text positions that should be changed and then decide how to change them. We constructed the new dataset WikiBioCTE for this task based on the existing dataset \textsc{WikiBio} (originally created for table-to-text generation). We use WikiBioCTE for training, and manually labeled a test set for testing. We also propose novel evaluation metrics and a novel method for solving the new task. Experimental results on the test set show that our proposed method is a good fit for this novel NLP task.

Controlling Text Edition by Changing Answers of Specific Questions

Recently, driven by numerous publicly available machine reading comprehension (MRC) datasets, MRC systems have made some progress. These datasets, however, have two major limitations: 1) the defined tasks are relatively simple, and 2) they do not provide explainable evaluation which is critical to objectively and comprehensively review the reasoning capabilities of current MRC systems. In this paper, we propose GCRC, a new dataset with challenging and high-quality multi-choice questions, collected from Gaokao Chinese (Chinese subject from the National College Entrance Examination of China). We have manually labelled three types of evidence to evaluate MRC systems’ reasoning process: 1) sentence-level relevant supporting facts in an article required for answering a given question, 2) error reason of a distractor (i.e., an incorrect option) for explaining why a distractor should be eliminated, which is an important reasoning step for multi-choice questions, and 3) types of reasoning skills required for answering questions. Extensive experiments show that our proposed dataset is more challenging and very useful for identifying the limitations of existing MRC systems in an explainable way, facilitating researchers to develop novel machine learning and reasoning approaches to tackle this challenging research problem.

GCRC: A New Challenging MRC Dataset from Gaokao Chinese for Explainable Evaluation

Identifying events and mapping them to a pre-defined taxonomy of event types has long been an important NLP problem. Most previous work has relied heavily on labor-intensive, domain-specific, annotation, ignoring the {\em semantic meaning} of the event types’ labels. Consequently, the learned models cannot effectively generalize to new label taxonomies and domains. We propose a zero-shot event extraction approach, which first identifies events with existing tools (e.g., SRL) and then maps them to a given taxonomy of event types in a zero-shot manner. Specifically, we leverage label representations induced by pre-trained language models, and map identified events to the target types via representation similarity. To semantically type the events’ arguments, we further use the definition of the events (e.g., argument of type ``Victim'' appears as the argument of event of type ``Attack'') as global constraints to regularize the prediction. The proposed approach is shown to be very effective on the ACE-2005 dataset, which has 33 trigger and 22 argument types. Without using any annotation, we successfully map 83% of the triggers and 54% of the arguments to the semantic correct types, almost doubling the performance of previous zero-shot approaches.

Using Adversarial Attacks to Reveal the Statistical Bias in Machine Reading Comprehension Models

Downloads

Next from ACL 2021

Zero-shot Event Extraction via Transfer Learning: Challenges and Insights

Similar lecture

DuReader_robust: A Chinese Dataset Towards Evaluating Robustness and Generalization of Machine Reading Comprehension in Real-World Applications

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Using Adversarial Attacks to Reveal the Statistical Bias in Machine Reading Comprehension Models

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2021

Zero-shot Event Extraction via Transfer Learning: Challenges and Insights

Similar lecture

DuReader_robust: A Chinese Dataset Towards Evaluating Robustness and Generalization of Machine Reading Comprehension in Real-World Applications

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads