Dominican Republic

Question Answering (QA) tasks requiring information from multiple documents often rely on a retrieval model to identify relevant information for reasoning. The retrieval model is typically trained to maximize the likelihood of the labeled supporting evidence. However, when retrieving from large text corpora such as Wikipedia, the correct answer can often be obtained from multiple evidence candidates. Moreover, not all such candidates are labeled as positive during annotation, rendering the training signal weak and noisy. This problem is exacerbated when the questions are unanswerable or when the answers are Boolean, since the model cannot rely on lexical overlap to make a connection between the answer and supporting evidence. We develop a new parameterization of set-valued retrieval that handles unanswerable queries, and we show that marginalizing over this set during training allows a model to mitigate false negatives in supporting evidence annotations. We test our method on two multi-document QA datasets, IIRC and HotpotQA. On IIRC, we show that joint modeling with marginalization improves model performance by 5.5 F1 points and achieves a new state-of-the-art performance of 50.5 F1. We also show that retrieval marginalization results in 4.1 QA F1 improvement over a non-marginalized baseline on HotpotQA in the fullwiki setting.

EMNLP 2021

Mitigating False-Negative Contexts in Multi-document Question Answering with Retrieval Marginalization

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible. The experience for on-site participants would closely approximate a normal pre-COVID *ACL conference, with 5-6 thematically organized parallel sessions and live Q/A and interactive discussion immediately after the talks. Presentations by virtual participants will be equitably interleaved with those of on-site participants, projected on the auditorium screens as if on-site, and also followed immediately by live Q/A and interactive discussion at a time during reasonable waking hours for the virtual presenter. For all participants, on-site and virtual, who are unable to attend a session due to either time-zone issues or because they are participating in another session live, talk recordings and slides will be available online at a minimum after the live presentation (and in many cases before as well), and questions may be submitted in advance on session-specific discussion boards and answered live in session with the usual visual aids if desired.

<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/ceby/"  seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

Please Note: The EMNLP registration system is not currently connected to the underline site as we are still in the process of building out EMNLP 2021. You will receive access instructions from underline the week of November 1st. 

Access is given only to EMNLP upon registration, if you have not registered please do so [here](https://2021.emnlp.org/registration).

Registered attendees will receive access the week of November 1st.

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible.

poster

Standard train-dev-test splits used to benchmark multiple models against each other are ubiquitously used in Natural Language Processing (NLP). In this setup, the train data is used for training the model, the development set for evaluating different versions of the proposed model(s) during development, and the test set to confirm the answers to the main research question(s). However, the introduction of neural networks in NLP has led to a different use of these standard splits; the development set is now often used for model selection during the training procedure. Because of this, comparing multiple versions of the same model during development leads to overestimation on the development data. As an effect, people have started to compare an increasing amount of models on the test data, leading to faster overfitting and ``expiration'' of our test sets. We propose to use a tune-set when developing neural network methods, which can be used for model picking so that comparing the different versions of a new model can safely be done on the development data.

We Need to Talk About train-dev-test Splits

Existing works have obvious deficiencies. (1) They usually consider only a single granularity of modification strategies (e.g. word-level or sentence-level), which is insufficient to explore the holistic textual space for generation; (2) They need to query victim models hundreds of times to make a successful attack, which is highly inefficient in practice. We propose MAYA, a Multi-grAnularitY Attack model to effectively generate high-quality adversarial samples with fewer queries to victim models. Furthermore, we propose a reinforcement-learning based method to train a multi-granularity attack agent through behavior cloning with the expert knowledge from our MAYA algorithm to further reduce the query times. Additionally, we also adapt the agent to attack black-box models that only output labels without confidence scores.

Multi-granularity Textual Adversarial Attack with Behavior Cloning

We introduce a high-quality and large-scale Vietnamese-English parallel dataset of 3.02M sentence pairs, which is 2.9M pairs larger than the benchmark Vietnamese-English machine translation corpus IWSLT15. We conduct experiments comparing strong neural baselines and well-known automatic translation engines on our dataset and find that in both automatic and human evaluations: the best performance is obtained by fine-tuning the pre-trained sequence-to-sequence denoising auto-encoder mBART. To our best knowledge, this is the first large-scale Vietnamese-English machine translation study. We hope our publicly available dataset and study can serve as a starting point for future research and applications on Vietnamese-English machine translation. We release our dataset at: https://github.com/VinAIResearch/PhoMT

PhoMT: A High-Quality and Large-Scale Benchmark Dataset for Vietnamese-English Machine Translation

Open-domain question answering has exploded in popularity recently due to the success of dense retrieval models, which have surpassed sparse models using only a few supervised training examples. However, in this paper, we demonstrate current dense models are not yet the holy grail of retrieval. We first construct EntityQuestions, a set of simple, entity-rich questions based on facts from Wikidata (e.g., "Where was Arve Furset born?"), and observe that dense retrievers drastically underperform sparse methods. We investigate this issue and uncover that dense retrievers can only generalize to common entities unless the question pattern is explicitly observed during training. We discuss two simple solutions towards addressing this critical problem. First, we demonstrate that data augmentation is unable to fix the generalization problem. Second, we argue a more robust passage encoder helps facilitate better question adaptation using specialized question encoders. We hope our work can shed light on the challenges in creating a robust, universal dense retriever that works well across different input distributions.

Simple Entity-Centric Questions Challenge Dense Retrievers

Although many studies use the LIWC lexicon to show the existence of verbal leakage cues in lie detection datasets, none mention how verbal leakage cues are influenced by means of data collection, or the impact thereof on the performance of models. In this paper, we study verbal leakage cues to understand the effect of the data construction method on their significance, and examine the relationship between such cues and models’ validity. The LIWC word-category dominance scores of seven lie detection datasets are used to show that audio statements and lie-based annotations indicate a greater number of strong verbal leakage cue categories. Moreover, we evaluate the validity of state-of-the-art lie detection models with cross- and in-dataset testing. Results show that in both types of testing, models trained on a dataset with more strong verbal leakage cue categories—as opposed to only a greater number of strong cues—yield superior results, suggesting that verbal leakage cues are a key factor for selecting lie detection datasets.

Lying Through One's Teeth: A Study on Verbal Leakage Cues

Knowledge-dependent tasks typically use two sources of knowledge: parametric, learned at training time, and contextual, given as a passage at inference time. To understand how models use these sources together, we formalize the problem of knowledge conflicts, where the contextual information contradicts the learned information. Analyzing the behaviour of popular models, we measure their over-reliance on memorized information (the cause of hallucinations), and uncover important factors that exacerbate this behaviour. Lastly, we propose a simple method to mitigate over-reliance on parametric knowledge, which minimizes hallucination, and improves out-of-distribution generalization by 4%-7%. Our findings demonstrate the importance for practitioners to evaluate model tendency to hallucinate rather than read, and show that our mitigation strategy encourages generalization to evolving information (i.e., time-dependent queries). To encourage these practices, we have released our framework for generating knowledge conflicts.

Entity-Based Knowledge Conflicts in Question Answering

One can find dozens of data resources for various languages in which coreference – a relation between two or more expressions that refer to the same real-world entity – is manually annotated. One could also assume that such expressions usually constitute syntactically meaningful units; however, mention spans have been annotated simply by delimiting token intervals in most coreference projects, i.e., independently of any syntactic representation. We argue that it could be advantageous to make syntactic and coreference annotations convergent in the long term. We present a pilot empirical study focused on matches and mismatches between hand-annotated linear mention spans and automatically parsed syntactic trees that follow Universal Dependencies conventions. The study covers 9 datasets for 8 different languages.

Do UD Trees Match Mention Spans in Coreference Annotations?

Recent works have found evidence of gender bias in models of machine translation and coreference resolution using mostly synthetic diagnostic datasets. While these quantify bias in a controlled experiment, they often do so on a small scale and consist mostly of artificial, out-of-distribution sentences. In this work, we find grammatical patterns indicating stereotypical and non-stereotypical gender-role assignments (e.g., female nurses versus male dancers) in corpora from three domains, resulting in a first large-scale gender bias dataset of 108K diverse real-world English sentences. We manually verify the quality of our corpus and use it to evaluate gender bias in various coreference resolution and machine translation models. We find that all tested models tend to over-rely on gender stereotypes when presented with natural inputs, which may be especially harmful when deployed in commercial systems. Finally, we show that our dataset lends itself to finetuning a coreference resolution model, finding it mitigates bias on a held out set. Our dataset and models are publicly available at github.com/SLAB-NLP/BUG. We hope they will spur future research into gender bias evaluation mitigation techniques in realistic settings.

Collecting a Large-Scale Gender Bias Dataset for Coreference Resolution and Machine Translation

Day-to-day and professional information needs are most often met only by combining information scattered across multiple texts. Accordingly, many NLP and text processing applications confront multi-text information, including multi-document summarization, multi-hop and conversational QA, knowledge base extraction, text mining and more. Properly performing these tasks requires identifying a range of informational links and relations across texts. While identifying such relations is a common infrastructure ingredient, there is very little application-independent foundational research in this area, where even the basic task of cross-document coreference resolution is heavily underexplored. In this talk I will present our line of work on modeling multi-text information. I will first describe infrastructure contributions to cross-document coreference resolution: a more realistic evaluation protocol, a corpus-wide (rather than topic-based) event coreference dataset, and an efficient crowdsourcing annotation tool. Next, I will describe a novel Cross-document Language Model (CDLM), which is geared to model cross-text information and hence better supports multi-text tasks. Lastly, I will present new tasks that extend the scope of cross-text informational links to be modeled. These include hierarchical cross-document coreference for scientific concepts, aligning matching proposition spans, and aligning propositional predicate-argument relations which are represented by QA-SRL question-answer pairs. I will conclude by discussing prospects for the evolution of this research line, while suggesting the added value of the multi-text setting as a foundational text understanding touchstone.

Modeling informational relations across multiple texts

The Fact Extraction and VERification Over Unstructured and Structured information (FEVEROUS) shared task, asks participating systems to determine whether human-authored claims are Supported or Refuted based on evidence retrieved from Wikipedia (or NotEnoughInfo if the claim cannot be verified). 
Compared to the FEVER 2018 shared task, the main challenge is the addition of structured data (tables and lists) as a source of evidence.
The claims in the FEVEROUS dataset can be verified using only structured evidence, only unstructured evidence, or a mixture of both. Submissions are evaluated using the FEVEROUS score that combines label accuracy and evidence retrieval. 
The shared task received 13 entries, six of which were able to beat the baseline system. The winning team was ``Bust a move!'', achieving a FEVEROUS score of 27% (+9% compared to the baseline). In this paper we describe the shared task, present the full results and highlight commonalities and innovations among the participating systems.

Mitigating False-Negative Contexts in Multi-document Question Answering with Retrieval Marginalization

Downloads

Next from EMNLP 2021

We Need to Talk About train-dev-test Splits

Similar lecture

How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answering