Croatia

A common limitation of diagnostic tests for detecting social biases in NLP models is that they may only detect stereotypic associations that are pre-specified by the designer of the test. Since enumerating all possible problematic associations is infeasible, it is likely these tests fail to detect biases that are present in a model but not pre-specified by the designer. To address this limitation, we propose SODAPOP (SOcial bias Discovery from Answers about PeOPle), an approach for automatic social bias discovery in social commonsense question-answering. The SODAPOP pipeline generates modified instances from the Social IQa dataset (Sap et al., 2019b) by (1) substituting names associated with different demographic groups, and (2) generating many distractor answers from a masked language model. By using a social commonsense model to score the generated distractors, we are able to uncover the model&#39;s stereotypic associations between demographic groups and an open set of words. We also test SODAPOP on debiased models and show the limitations of multiple state-of-the-art debiasing algorithms.

EACL 2023

SODAPOP: Open-Ended Discovery of Social Biases in Social Commonsense Reasoning Models

first-name biases

social commonsense reasoning

fairness and bias

bias detection

social bias

A common limitation of diagnostic tests for detecting social biases in NLP models is that they may only detect stereotypic associations that are pre-specified by the designer of the test. Since enumerating all possible problematic associations is infeasible, it is likely these tests fail to detect biases that are present in a model but not pre-specified by the designer. To address this limitation, we propose SODAPOP (SOcial bias Discovery from Answers about PeOPle), an approach for automatic social bias discovery in social commonsense question-answering. The SODAPOP pipeline generates modified instances from the Social IQa dataset (Sap et al., 2019b) by (1) substituting names associated with different demographic groups, and (2) generating many distractor answers from a masked language model. By using a social commonsense model to score the generated distractors, we are able to uncover the model's stereotypic associations between demographic groups and an open set of words. We also test SODAPOP on debiased models and show the limitations of multiple state-of-the-art debiasing algorithms.

poster

### Welcome to the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL) 
Continuing its mission of expanding and involving the science community of all European countries, EACL had selected the Ukrainian community for the 16th EACL, which took place online due to the COVID pandemic. Unfortunately, the ongoing war made the organisation in Kyiv impossible. Considering the importance of physical interaction among researchers, especially after the restrictions imposed by the COVID pandemic, in addition to an online mode, the conference will be held in Dubrovnik, Croatia, from 2 to 6 of May, 2023. The original aim of strengthening the connection with the Ukrainian community will not change as our program will feature a dedicated session and a workshop to highlight work on Ukrainian language technologies. As the flagship European conference in the field of computational linguistics, EACL welcomes European and international researchers covering a broad spectrum of research areas that are concerned with computational approaches to natural language.

You need to register for the conference in order to access this site. Please visit https://2023.eacl.org/registration for more information.

As the flagship European conference in the field of computational linguistics, EACL welcomes European and international researchers covering a broad spectrum of research areas that are concerned with computational approaches to natural language.

Event coreference models cluster event mentions pertaining to the same real-world event. Recent models rely on contextualized representations to recognize coreference among lexically or contextually similar mentions. However, models typically fail to leverage commonsense inferences, which is particularly limiting for resolving lexically-divergent mentions. We propose a model that extends event mentions with temporal commonsense inferences. Given a complex sentence with multiple events, e.g., "the man killed his wife and got arrested'', with the target event "arrested'', our model generates plausible events that happen before the target event – such as "the police arrived'', and after it, such as "he was sentenced''. We show that incorporating such inferences into an existing event coreference model improves its performance, and we analyze the coreferences in which such temporal knowledge is required.

What happens before and after: Multi-Event Commonsense in Event Coreference Resolution

In task-oriented dialog (ToD) new intents emerge on regular basis, with a handful of available utterances at best. This renders effective Few-Shot Intent Classification (FSIC) a central challenge for modular ToD systems. Recent FSIC methods appear to be similar: they use pretrained language models (PLMs) to encode utterances and predominantly resort to nearest-neighbor-based inference. However, they also differ in major components: they start from different PLMs, use different encoding architectures and utterance similarity functions, and adopt different training regimes.Coupling of these vital components together with the lack of informative ablations prevents the identification of factors that drive the (reported) FSIC performance. We propose a unified framework to evaluate these components along the following key dimensions:(1) Encoding architectures: Cross-Encoder vs Bi-Encoders;
(2) Similarity function: Parameterized (i.e., trainable) vs non-parameterized; 
(3) Training regimes: Episodic meta-learning vs conventional (i.e., non-episodic) training. Our experimental results on seven FSIC benchmarks reveal three new important findings. First, the unexplored combination of cross-encoder architecture and episodic meta-learning consistently yields the best FSIC performance. Second, episodic training substantially outperforms its non-episodic counterpart. Finally, we show that splitting episodes into support and query sets has a limited and inconsistent effect on performance. Our findings show the importance of ablations and fair comparisons in FSIC. We publicly release our code and data.

The Devil is in the Details: On Models and Training Regimes for Few-Shot Intent Classification

Recent years have witnessed interest in Temporal Question Answering over Knowledge Graphs (TKGQA), resulting in the development of multiple methods. However, these are highly engineered, thereby limiting their generalizability, and they do not automatically discover relevant parts of the KG during multi-hop reasoning. Relational graph convolutional networks (RGCN) provide an opportunity to address both of these challenges -- we explore this direction in the paper. Specifically, we propose a novel, intuitive and interpretable scheme to modulate the messages passed through a KG edge during convolution based on the relevance of its associated period to the question. We also introduce a gating device to predict if the answer to a complex temporal question is likely to be a KG entity or time and use this prediction to guide our scoring mechanism. We evaluate the resulting system, which we call TwiRGCN, on a recent challenging dataset for multi-hop complex temporal QA called TimeQuestions. We show that TwiRGCN significantly outperforms state-of-the-art models on this dataset across diverse question types. Interestingly, TwiRGCN improves accuracy by 9--10 percentage points for the most difficult ordinal and implicit question types.

TwiRGCN: Temporally Weighted Graph Convolution for Question Answering over Temporal Knowledge Graphs

The most popular approach in grammatical error correction (GEC) is based on sequence-to-sequence (seq2seq) models. Similar to other autoregressive generation tasks, seq2seq GEC also faces the exposure bias problem, i.e., the context tokens are drawn from different distributions during training and testing, caused by the teacher forcing mechanism. In this paper, we propose a novel data manipulation approach to overcome this problem, which includes a data augmentation method during training to mimic the decoder input at inference time, and a data reweighting method to automatically balance the importance of each kind of augmented samples. Experimental results on benchmark GEC datasets show that our method achieves significant improvements compared to prior approaches.

Mitigating Exposure Bias in Grammatical Error Correction with Data Augmentation and Reweighting

In this work, we focus on low-resource dependency parsing for multiple languages. Several strategies are tailored to enhance performance in low-resource scenarios. While these are well-known to the community, it is not trivial to select the best-performing combination of these strategies for a low-resource language that we are interested in, and not much attention has been given to measuring the efficacy of these strategies. We experiment with 5 low-resource strategies for our ensembled approach on 7 Universal Dependency (UD) low-resource languages. Our exhaustive experimentation on these languages supports the effective improvements for languages not covered in pretrained models. We show a successful application of the ensembled system on a truly low-resource language Sanskrit. The code and data are available at: this https URL

Systematic Investigation of Strategies Tailored for Low-Resource Settings for Low-Resource Dependency Parsing

The dissemination of false information on the internet has received considerable attention over the last decade. Misinformation often spreads faster than mainstream news, thus making manual fact checking inefficient or, at best, labor-intensive. Therefore, there is an increasing need to develop methods for automatic detection of misinformation. Although resources for creating such methods are available in English, other languages are often under-represented in this effort.
With this contribution, we present IRMA, a corpus containing over 600,000 Italian news articles (335+ million tokens) collected from 56 websites classified as `untrustworthy' by professional fact-checkers. The corpus is freely available and comprises a rich set of text- and website-level data, representing a turnkey resource to test hypotheses and develop automatic detection algorithms. 
It contains texts, titles, and dates (from 2004 to 2022), along with three types of semantic measures (i.e., keywords, topics at three different resolutions, and LIWC lexical features). 
IRMA also includes domain-specific information such as source type (e.g., political, health, conspiracy, etc.), credibility, and higher-level metadata, including several metrics of website incoming traffic that allow to investigate user online behavior. 
IRMA constitutes the largest corpus of misinformation available today in Italian, making it a valid tool for advancing quantitative research on untrustworthy news detection and ultimately helping limit the spread of misinformation.

IRMA: the 335-million-word Italian coRpus for studying MisinformAtion

Subjective bias is ubiquitous on news sites, social media, and knowledge resources like Wikipedia. Many existing methods for subjective bias correction have typically focused on making one-word edits and have been trained over a single (often, noisy) domain. In contrast, we propose a novel reinforced sequence training approach for robust subjective bias correction. Three of the unique characteristics of the approach are: (i) it balances bias neutralization with fluency and semantics preservation through reinforcement learning, to broaden the scope to bias beyond a single word; (ii) it is cross-trained over multiple sources of bias to be more robust to new styles of biased writing that are not seen in the training data for a single domain; and (iii) it is used to fine-tune a large pre-trained transformer model to yield state-of-the-art performance in bias text correction task. Extensive experiments show that the proposed approach results in significant improvements in subjective bias correction versus alternatives.

Reinforced Sequence Training based Subjective Bias Correction

Language contact is a pervasive phenomenon reflected in the borrowing of words from donor to recipient languages. Most computational approaches to borrowing detection treat all languages under study as equally important, even though dominant languages have a stronger impact on heritage languages than vice versa. We test new methods for lexical borrowing detection in contact situations where dominant languages play an important role, applying two classical sequence comparison methods and one machine learning method to a sample of seven Latin American languages which have all borrowed extensively from Spanish. All methods perform well, with the supervised machine learning system outperforming the classical systems. A review of detection errors shows that borrowing detection could be substantially improved by taking into account donor words with divergent meanings from recipient words.

Detecting Lexical Borrowings from Dominant Languages in Multilingual Wordlists

The RVL-CDIP benchmark is widely used for measuring performance on the task of document classification. Despite its widespread use, we reveal several undesirable characteristics of the RVL-CDIP benchmark. These include (1) substantial amounts of label noise, which we estimate to be 8.1% (ranging between 1.6% to 16.9% per document category); (2) presence of many ambiguous or multi-label documents; (3) a large overlap between test and train splits, which can inflate model performance metrics; and (4) presence of sensitive personally-identifiable information like US Social Security numbers (SSNs). We argue that there is a risk in using RVL-CDIP for benchmarking document classifiers, as its limited scope, presence of errors (state-of-the-art models now achieve accuracy error rates that are within our estimated label error rate), and lack of diversity make it less than ideal for benchmarking. We further advocate for the creation of a new document classification benchmark, and provide recommendations for what characteristics such a resource should include.

On Evaluation of Document Classifiers using RVL-CDIP

Comprehending an article requires understanding its constituent events. However, the context where an event is mentioned often lacks the details of this event. A question arises: how can the reader obtain more knowledge about this particular event in addition to what is provided by the local context in the article?
This work defines Event Linking, a new natural language understanding task at the event level. Event linking tries to link an event mention appearing in an article to the most appropriate Wikipedia page. This page is expected to provide rich knowledge about what the event mention refers to. To standardize the research in this new direction, we contribute in fourfold. First, this is the first work in the community that formally defines the Event Linking task. Second, we collect a dataset for this new task. Specifically, we automatically gather the training set from Wikipedia, and then create two evaluation sets: one from the Wikipedia domain, reporting the in-domain performance, and a second from the real-world news domain, to evaluate out-of-domain performance. Third, we retrain and evaluate two state-of-the-art (SOTA) entity linking models, showing the challenges of event linking, and we propose an
event-specific linking system, EVELINK, to set a competitive result for the new task. Fourth, we conduct a detailed and insightful analysis to help understand the task and the limitations of the current model. Overall, as our analysis shows, Event Linking is a challenging and essential task requiring more effort from the community.

Downloads

Next from EACL 2023

What happens before and after: Multi-Event Commonsense in Event Coreference Resolution