Dominican Republic

Building compositional explanations requires models to combine two or more facts that, together, describe why the answer to a question is correct. Typically, these &quot;multi-hop&quot; explanations are evaluated relative to one (or a small number of) gold explanations. In this work, we show these evaluations substantially underestimate model performance, both in terms of the relevance of included facts, as well as the completeness of model-generated explanations, because models regularly discover and produce valid explanations that are different than gold explanations. To address this, we construct a large corpus of 126k domain-expert (science teacher) relevance ratings that augment a corpus of explanations to standardized science exam questions, discovering 80k additional relevant facts not rated as gold. We build three strong models based on different methodologies (generation, ranking, and schemas), and empirically show that while expert-augmented ratings provide better estimates of explanation quality, both original (gold) and expert-augmented automatic evaluations still substantially underestimate performance by up to 36% when compared with full manual expert judgements, with different models being disproportionately affected. This poses a significant methodological challenge to accurately evaluating explanations produced by compositional reasoning models.

EMNLP 2021

On the Challenges of Evaluating Compositional Explanations in Multi-Hop Inference: Relevance, Completeness, and Expert Ratings

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.

Building compositional explanations requires models to combine two or more facts that, together, describe why the answer to a question is correct. Typically, these "multi-hop" explanations are evaluated relative to one (or a small number of) gold explanations. In this work, we show these evaluations substantially underestimate model performance, both in terms of the relevance of included facts, as well as the completeness of model-generated explanations, because models regularly discover and produce valid explanations that are different than gold explanations. To address this, we construct a large corpus of 126k domain-expert (science teacher) relevance ratings that augment a corpus of explanations to standardized science exam questions, discovering 80k additional relevant facts not rated as gold. We build three strong models based on different methodologies (generation, ranking, and schemas), and empirically show that while expert-augmented ratings provide better estimates of explanation quality, both original (gold) and expert-augmented automatic evaluations still substantially underestimate performance by up to 36% when compared with full manual expert judgements, with different models being disproportionately affected. This poses a significant methodological challenge to accurately evaluating explanations produced by compositional reasoning models.

poster

The factual knowledge acquired during pre-training and stored in the parameters of Language Models (LMs) can be useful in downstream tasks (e.g., question answering or textual inference). However, some facts can be incorrectly induced or become obsolete over time. We present KnowledgeEditor, a method which can be used to edit this knowledge and, thus, fix 'bugs' or unexpected predictions without the need for expensive re-training or fine-tuning. Besides being computationally efficient, KnowledgeEditordoes not require any modifications in LM pre-training (e.g., the use of meta-learning). In our approach, we train a hyper-network with constrained optimization to modify a fact without affecting the rest of the knowledge; the trained hyper-network is then used to predict the weight update at test time. We show KnowledgeEditor's efficacy with two popular architectures and knowledge-intensive tasks: i) a BERT model fine-tuned for fact-checking, and ii) a sequence-to-sequence BART model for question answering. With our method, changing a prediction on the specific wording of a query tends to result in a consistent change in predictions also for its paraphrases. We show that this can be further encouraged by exploiting (e.g., automatically-generated) paraphrases during training. Interestingly, our hyper-network can be regarded as a 'probe' revealing which components need to be changed to manipulate factual knowledge; our analysis shows that the updates tend to be concentrated on a small subset of components. Source code available a https://github.com/nicola-decao/KnowledgeEditor

Editing Factual Knowledge in Language Models

We propose a new task and dataset, MultiDoc2Dial, on modeling goal-oriented dialogues grounded in multiple documents. Most previous works treat document-grounded dialogue modeling as a machine reading comprehension task based on a single given document or passage. In this work, we aim to address more realistic scenarios where a goal-oriented information-seeking conversation involves multiple topics, and hence is grounded on different documents. To facilitate such a task, we introduce a new dataset that contains dialogues grounded in multiple documents from four different domains. We also explore modeling the dialogue-based and document-based context in the dataset. We present strong baseline approaches and various experimental results, aiming to support further research efforts on such a task.

MultiDoc2Dial: Modeling Dialogues Grounded in Multiple Documents

Text generation systems are ubiquitous in natural language processing applications. However, evaluation of these systems remains a challenge, especially in multilingual settings. In this paper, we propose L'AMBRE -- a metric to evaluate the morphosyntactic well-formedness of text using its dependency parse and morphosyntactic rules of the language. We present a way to automatically extract various rules governing morphosyntax directly from dependency treebanks. To tackle the noisy outputs from text generation systems, we propose a simple methodology to train robust parsers. We show the effectiveness of our metric on the task of machine translation through a diachronic study of systems translating into morphologically-rich languages.

Evaluating the Morphosyntactic Well-formedness of Generated Texts

Current NLP models are predominantly trained through a two-stage "pre-train then fine-tune" pipeline. Prior work has shown that inserting an intermediate pre-training stage, using heuristic masking policies for masked language modeling (MLM), can significantly improve final performance. However, it is still unclear (1) in what cases such intermediate pre-training is helpful, (2) whether hand-crafted heuristic objectives are optimal for a given task, and (3) whether a masking policy designed for one task is generalizable beyond that task. In this paper, we perform a large-scale empirical study to investigate the effect of various masking policies in intermediate pre-training with nine selected tasks across three categories. Crucially, we introduce methods to automate the discovery of optimal masking policies via direct supervision or meta-learning. We conclude that the success of intermediate pre-training is dependent on appropriate pre-train corpus, selection of output format (i.e., masked spans or full sentence), and clear understanding of the role that MLM plays for the downstream task. In addition, we find our learned masking policies outperform the heuristic of masking named entities on TriviaQA, and policies learned from one task can positively transfer to other tasks in certain cases, inviting future research in this direction.

On the Influence of Masking Policies in Intermediate Pre-training

Many recent studies have demonstrated that unsupervised sentence representations of neural networks encode syntactic information by observing that neural language models are able to predict the agreement between a verb and its subject. We take a critical look at this line of research by showing that it is possible to achieve high accuracy on this agreement task with simple surface heuristics, indicating a possible flaw in our assessment of neural networks' syntactic ability. Our fine-grained analyses of results on the long-range French object-verb agreement show that contrary to LSTMs, Transformers are able to capture a non-trivial amount of grammatical structure.

Are Transformers a Modern Version of ELIZA? Observations on French Object Verb Agreement

Automatic fact verification has attracted recent research attention as the increasing dissemination of disinformation on social media platforms. The FEVEROUS shared task introduces a benchmark for fact verification, in which a system is challenged to verify the given claim using the extracted evidential elements from Wikipedia documents. In this paper, we propose our 3rd place three-stage system consisting of document retrieval, element retrieval, and verdict inference for the FEVEROUS shared task. By considering both the word frequency and semantic context in the fact extraction and verification task, our system achieves 0.29 FEVEROUS score on the development set and 0.25 FEVEROUS score on the blind test set, both outperforming the FEVEROUS baseline.

NCU: Verdict Inference with Claim and Retrieved Elements Using RoBERTa

The widespread use of the Internet and the rapid dissemination of information poses the challenge of identifying the veracity of its content. Stance detection, which is the task of predicting the position of a text in regard to a specific target (e.g. claim or debate question), has been used to determine the veracity of information in tasks such as rumor classification and fake news detection. While most of the work and available datasets for stance detection address short texts snippets extracted from textual dialogues, social media platforms, or news headlines with a strong focus on the English language, there is a lack of resources targeting long texts in other languages. Our contribution in this paper is twofold. First, we present a German dataset of debate questions and news articles that is manually annotated for stance and emotion detection. Second, we leverage the dataset to tackle the supervised task of classifying the stance of a news article with regards to a debate question and provide baseline models as a reference for future work on stance detection in German news articles.

Stance Detection in German News Articles

As part of the FEVEROUS shared task, we developed a robust and finely tuned architecture to handle the joint retrieval and entailment on text data as well as structured data like tables. We proposed two training schemes to tackle the hurdles inherent to multi-hop multi-modal datasets. The first one allows having a robust retrieval of full evidence sets, while the second one enables entailment to take full advantage of noisy evidence inputs. In addition, our work has revealed important insights and potential avenue of research for future improvement on this kind of dataset. In preliminary evaluation on the FEVEROUS shared task test set, our system achieves 0.271 FEVEROUS score, with 0.4258 evidence recall and 0.5607 entailment accuracy.

FaBULOUS: Fact-checking Based on Understanding of Language Over Unstructured and Structured information

Dialogue systems powered by large pretrained language models exhibit an innate
ability to deliver fluent and natural-sounding responses. Despite their impressive performance, these models are fitful and can often generate factually incorrect statements impeding their widespread adoption. In this paper,
we focus on the task of improving faithfulness and reducing hallucination of neural dialogue systems to known facts supplied by a Knowledge Graph (KG). We propose NEURAL PATH HUNTER which follows a generate-then- refine strategy whereby a generated response is amended using the KG. NEURAL PATH HUNTER leverages a separate token-level fact critic to identify plausible sources of hallucination followed by a refinement stage that retrieves correct entities by crafting a query signal that is propagated over a k-hop subgraph. We empirically validate our proposed
approach on the OpenDialKG dataset (Moon et al., 2019) against a suite of metrics
and report a relative improvement of faithfulness over dialogue responses by 20:35% based on FeQA (Durmus et al., 2020).

Neural Path Hunter: Reducing Hallucination in Dialogue Systems via Path Grounding

Practical dialogue systems require robust methods of detecting out-of-scope (OOS) utterances to avoid conversational breakdowns and related failure modes. Directly training a model with labeled OOS examples yields reasonable performance, but obtaining such data is a resource-intensive process. To tackle this limited-data problem, previous methods focus on better modeling the distribution of in-scope (INS) examples. We introduce GOLD as an orthogonal technique that augments existing data to train better OOS detectors operating in low-data regimes. GOLD generates pseudo-labeled candidates using samples from an auxiliary dataset and keeps only the most beneficial candidates for training through a novel filtering mechanism. In experiments across three target benchmarks, the top GOLD model outperforms all existing methods on all key metrics: AUROC, AUPR, FPR@N. We also analyze the unique properties of OOS data to identify key factors for optimally applying our proposed method.

Premium content

Downloads

Next from EMNLP 2021

Editing Factual Knowledge in Language Models

Similar lecture

Evaluating the Evaluation Metrics for Style Transfer: A Case Study in Multilingual Formality Transfer

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES