Croatia

Existing state-of-the-art methods for open-domain question-answering (ODQA) use an open book approach in which information is first retrieved from a large text corpus or knowledge base (KB) and then reasoned over to produce an answer. A recent alternative is to retrieve from a collection of previously-generated question-answer pairs; this has several practical advantages including being more memory and compute-efficient. Question-answer pairs are also appealing in that they can be viewed as an intermediate between text and KB triples: like KB triples, they often concisely express a single relationship, but like text, have much higher coverage than traditional KBs. In this work, we describe a new QA system that augments a text-to-text model with a large memory of question-answer pairs, and a new pre-training task for the latent step of question retrieval. The pre-training task substantially simplifies training and greatly improves performance on smaller QA benchmarks. Unlike prior systems of this sort, our QA system can also answer multi-hop questions that do not explicitly appear in the collection of stored question-answer pairs.

EACL 2023

Augmenting Pre-trained Language Models with QA-Memory for Open-Domain Question Answering

memory augmentation

language model

retrieval

### 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.

poster

We describe a simple and effective method (Spectral Attribute removaL; SAL) to remove private or guarded information from neural representations. Our method uses matrix decomposition to project the input representations into directions with reduced covariance with the guarded information rather than maximal covariance as factorization methods normally use. We begin with linear information removal and proceed to generalize our algorithm to the case of nonlinear information removal using kernels. Our experiments demonstrate that our algorithm retains better main task performance after removing the guarded information compared to previous work. In addition, our experiments demonstrate that we need a relatively small amount of guarded attribute data to remove information about these attributes, which lowers the exposure to sensitive data and is more suitable for low-resource scenarios.

Gold Doesn't Always Glitter: Spectral Removal of Linear and Nonlinear Guarded Attribute Information

Dropout is a widely used regularization trick to resolve the overfitting issue in large feedforward neural networks trained on a small dataset, which performs poorly on the held-out test subset. Although the effectiveness of this regularization trick has been extensively studied for convolutional neural networks, there is a lack of analysis of it for unsupervised models and in particular, VAE-based neural topic models. In this paper, we have analyzed the consequences of dropout in the encoder as well as in the decoder of the VAE architecture in three widely used neural topic models, namely, contextualized topic model (CTM), ProdLDA, and embedded topic model (ETM) using four publicly available datasets. We characterize the dropout effect on these models in terms of the quality and predictive performance of the generated topics.

Do Neural Topic Models Really Need Dropout? Analysis of the Effect of Dropout in Topic Modeling

We propose two methods to make unsupervised domain adaptation (UDA) more parameter efficient using adapters -- small bottleneck layers interspersed with every layer of the large-scale pre-trained language model (PLM). The first method deconstructs UDA into a two-step process: first by adding a domain adapter to learn domain-invariant information and then by adding a task adapter that uses domain-invariant information to learn task representations in the source domain. The second method jointly learns a supervised classifier while reducing the divergence measure. 
Compared to strong baselines, our simple methods perform well in natural language inference (MNLI) and the cross-domain sentiment classification task. We even outperform unsupervised domain adaptation methods such as DANN and DSN in sentiment classification, and we are within 0.85\% F1 for natural language inference task, by fine-tuning only a fraction of the full model parameters. We release our code at this URL.

UDAPTER - Efficient Domain Adaptation Using Adapters

For end-to-end speech translation, regularizing the encoder with the Connectionist Temporal Classification (CTC) objective using the source transcript or target translation as labels can greatly improve quality. However, CTC demands an extra prediction layer over the vocabulary space, bringing in non-negligible model parameters and computational overheads, although this layer becomes useless at inference. In this paper, we re-examine the need for genuine vocabulary labels for CTC for regularization and explore strategies to reduce the CTC label space, targeting improved efficiency without quality degradation. We propose coarse labeling for CTC (CoLaCTC), which merges vocabulary labels via simple heuristic rules, such as using truncation, division or modulo (MOD) operations. Despite its simplicity, our experiments on 4 source and 8 target languages show that CoLaCTC with MOD particularly can compress the label space aggressively to 256 and even further, gaining training efficiency (1.18× ∼ 1.77× speedup depending on the original vocabulary size) yet still delivering comparable or better performance than the CTC baseline. We also show that CoLaCTC successfully generalizes to CTC regularization regardless of using transcript or translation for labeling.

Efficient CTC Regularization via Coarse Labels for End-to-End Speech Translation

We consider the problem of identifying a \textbackslash{}emph\{minimal subset\} of training data \$\textbackslash{}mathcal\{S\}\_t\$ such that if the instances comprising \$\textbackslash{}mathcal\{S\}\_t\$ had been removed prior to training, the categorization of a given test point \$x\_t\$ would have been different.
Identifying such a set may be of interest for a few reasons.
First, the cardinality of \$\textbackslash{}mathcal\{S\}\_t\$ provides a measure of robustness (if \$|\textbackslash{}mathcal\{S\}\_t|\$ is small for \$x\_t\$, we might be less confident in the corresponding prediction), which we show is correlated with but complementary to predicted probabilities.
Second, interrogation of \$\textbackslash{}mathcal\{S\}\_t\$ may provide a novel mechanism for \textbackslash{}emph\{contesting\} a particular model prediction: If one can make the case that the points in \$\textbackslash{}mathcal\{S\}\_t\$ are wrongly labeled or irrelevant, this may argue for overturning the associated prediction. 
Identifying \$\textbackslash{}mathcal\{S\}\_t\$ via brute-force is intractable.
We propose comparatively fast approximation methods to find \$\textbackslash{}mathcal\{S\}\_t\$ based on \textbackslash{}emph\{influence functions\}, and find that---for simple convex text classification models---these approaches can often successfully identify relatively small sets of training examples which, if removed, would flip the prediction.

How Many and Which Training Points Would Need to be Removed to Flip this Prediction?

Training machine learning models to successfully perform scientific fact-checking tasks is challenging due to the expertise bottleneck that limits the availability of appropriate training datasets. In this task, models use textual evidence to confirm scientific claims, which requires data that contains extensive domain-expert annotation. Consequently, the number of existing scientific-fact-checking datasets and the sizes of those datasets are limited. However, these limitations do not apply to multiple-choice question datasets because of the necessity of domain exams in the modern education system. As one of the first steps towards addressing the fact-checking dataset scarcity problem in scientific domains, we propose a pipeline for automatically converting multiple-choice questions into fact-checking data, which we call Multi2Claim. By applying the proposed pipeline, we generated two large-scale datasets for scientific-fact-checking tasks: Med-Fact and Gsci-Fact for the medical and general science domains, respectively. These two datasets are among the first examples of large-scale scientific-fact-checking datasets. We developed baseline models for the verdict prediction task using each dataset. Additionally, we demonstrated that the datasets could be used to improve performance with respect to the F 1 weighted metric on existing fact-checking datasets such as SciFact, HEALTHVER, COVID-Fact, and CLIMATE-FEVER. In some cases, the improvement in performance was up to a 26\% increase.

Multi2Claim: Generating Scientific Claims from Multi-Choice Questions for Scientific Fact-Checking

Prompting pre-trained language models leads to promising results across natural language processing tasks but is less effective when applied in low-resource domains, due to the domain gap between the pre-training data and the downstream task. In this work, we bridge this gap with a novel and lightweight prompting methodology called SwitchPrompt for the adaptation of language models trained on datasets from the general domain to diverse low-resource domains. Using domain-specific keywords with a trainable gated prompt, SwitchPrompt offers domain-oriented prompting, that is, effective guidance on the target domains for general-domain language models. Our few-shot experiments on three text classification benchmarks demonstrate the efficacy of the general-domain pre-trained language models when used with SwitchPrompt. They often even outperform their domain-specific counterparts trained with baseline state-of-the-art prompting methods by up to 10.7\% performance increase in accuracy. This result indicates that SwitchPrompt effectively reduces the need for domain-specific language model pre-training.

SwitchPrompt: Learning Domain-Specific Gated Soft Prompts for Classification in Low-Resource Domains

Large pre-trained language models contain societal biases and carry along these biases to downstream tasks. Current in-processing bias mitigation approaches (like adversarial training) impose debiasing by updating a model's parameters, effectively transferring the model to a new, irreversible debiased state. In this work, we propose a novel approach to develop stand-alone debiasing functionalities separate from the model, which can be integrated into the model on-demand, while keeping the core model untouched. Drawing from the concept of AdapterFusion in multi-task learning, we introduce DAM (Debiasing with Adapter Modules) – a debiasing approach to first encapsulate arbitrary bias mitigation functionalities into separate adapters, and then add them to the model on-demand in order to deliver fairness qualities. We conduct a large set of experiments on three classification tasks with gender, race, and age as protected attributes. Our results show that DAM improves or maintains the effectiveness of bias mitigation, avoids catastrophic forgetting in a multi-attribute scenario, and maintains on-par task performance, while granting parameter-efficiency and easy switching between the original and debiased models.

Parameter-efficient Modularised Bias Mitigation via AdapterFusion

Response generation is one of the critical components in task-oriented dialog systems. Existing studies have shown that large pre-trained language models can be adapted to this task. The typical paradigm of adapting such extremely large language models would be by fine-tuning on the downstream tasks which is not only time-consuming but also involves significant resources and access to fine-tuning data. Prompting (Schick and Schütze, 2020) has been an alternative to fine-tuning in many NLP tasks. In our work, we explore the idea of using prompting for response generation in task-oriented dialog systems. Specifically, we propose an approach that performs contextual dynamic prompting where the prompts are learnt from dialog contexts. We aim to distill useful prompting signals from the dialog context. On experiments with MultiWOZ 2.2 dataset (Zang et al., 2020), we show that contextual dynamic prompts improve response generation in terms of combined score (Mehri et al., 2019) by 3 absolute points, and an additional 17 points when dialog states are incorporated. Furthermore, we carried out human annotation on these conversations and found that agents which incorporate context are preferred over agents with vanilla prefix-tuning.

Contextual Dynamic Prompting for Response Generation in Task-oriented Dialog Systems

One significant obstacle to the successful application of machine learning to real-world data is that of labeling: it is often prohibitively expensive to pay an ethical amount for the human labor required to label a dataset successfully. Human-in-the-loop techniques such as active learning can reduce the cost, but the required human time is still significant and many fixed costs remain. Another option is to employ pre-trained transformer models as labelers at scale, which can yield reasonable accuracy and significant cost savings. However, such models can still be expensive to use due to their high computational requirements, and the opaque nature of these models is not always suitable in applied social science and public use contexts.

We propose a novel semi-supervised method, named Slingshot Learning, in which we iteratively and selectively augment a small human-labeled dataset with labels from a high-quality "teacher" model to slingshot the performance of a "student" model in a cost-efficient manner. This reduces the accuracy trade-off required to use these simpler algorithms without disrupting their benefits, such as lower compute requirements, better interpretability, and faster inference. We define and discuss the slingshot learning algorithm and demonstrate its effectiveness on several benchmark tasks, using ALBERT to teach a simple Naive Bayes binary classifier. We experimentally demonstrate that Slingshot learning effectively decreases the performance gap between the teacher and student models. We also analyze its performance in several scenarios and compare different variants of the algorithm.

Downloads

Next from EACL 2023

Gold Doesn't Always Glitter: Spectral Removal of Linear and Nonlinear Guarded Attribute Information

Similar lecture

Question-Answer Sentence Graph for Joint Modeling Answer Selection