Malta

While the Large Language Models (LLMs) dominate a majority of language understanding tasks, previous work shows that some of these results are supported by modelling spurious correlations of training datasets. Authors commonly assess model robustness by evaluating their models on out-of-distribution (OOD) datasets of the same task, but these datasets might share the bias of the training dataset. We propose a simple method for measuring a scale of models&#39; reliance on any identified spurious feature and assess the robustness towards a large set of known and newly found prediction biases for various pre-trained models and debiasing methods in Question Answering (QA). We find that the reported OOD gains of debiasing methods can not be explained by mitigated reliance on biased features, suggesting that biases are shared among different QA datasets. We further evidence this by measuring that performance of OOD models depends on bias features comparably to the ID model. Our findings motivate future work to refine the reports of LLMs&#39; robustness to a level of known spurious features.

EACL 2024

Think Twice: Measuring the Efficiency of Eliminating Prediction Shortcuts of Question Answering Models

prediction shortcuts

datasets

bias

question answering

generalization

evaluation

robustness

[![](https://assets.underline.io/uploads/markdown_image/1/image/81a3c0317d24f663b49b996875024d45.png)](https://aclanthology.org/volumes/2022.coling-1/)



### THE CONFERENCE WE KNOW AND WE WANT.
**COLING**, the International Conference on Computational Linguistics, is one of the premier conferences for the natural language processing and computational linguistics.

First established in 1965, the biennial COLING conference is held in diverse parts of the globe and attracts participants from both top-ranked research centers and emerging countries. Today, the most important developments in our field are taking place not only in universities and academic research institutes but also in industrial research departments including tech-startups. COLING provides opportunities for all these communities to showcase their exciting discovery.

In fall of 2022, COLING will be held in Gyeongju in a hybrid format. All participants can either present at the venue site or join virtually. As more people get vaccinated, we are happy to provide safer environments for our colleagues. We believe that COLING 2022 will be one of the conferences, free from the pandemic. The hybrid format gives presenters and sponsors a valuable opportunity to promote their companies in both an online and in-person venue. For the first time in a long time, customers can interact with their sponsor's products first-hand. The online venue, too, gives sponsors the chance to network with those unable to attend the in-person session.

The hybrid format ultimately lends itself to greater exposure for our sponsors: COLING2022 will let you reach more potential partners and customers than ever before!

To gain access to this event page you are required to register. Find more information and pay the registration fee on event the organizer’s website **[https://coling2022.org/reg](https://coling2022.org/reg)**

COLING 2022

COLING, the International Conference on Computational Linguistics, is one of the premier conferences for the natural language processing and computational linguistics.

While the Large Language Models (LLMs) dominate a majority of language understanding tasks, previous work shows that some of these results are supported by modelling spurious correlations of training datasets. Authors commonly assess model robustness by evaluating their models on out-of-distribution (OOD) datasets of the same task, but these datasets might share the bias of the training dataset. We propose a simple method for measuring a scale of models' reliance on any identified spurious feature and assess the robustness towards a large set of known and newly found prediction biases for various pre-trained models and debiasing methods in Question Answering (QA). We find that the reported OOD gains of debiasing methods can not be explained by mitigated reliance on biased features, suggesting that biases are shared among different QA datasets. We further evidence this by measuring that performance of OOD models depends on bias features comparably to the ID model. Our findings motivate future work to refine the reports of LLMs' robustness to a level of known spurious features.

technical paper

Welcome to the 18th 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 has selected the Malta community for the 18th EACL. Considering the importance of physical interaction among researchers, the conference will be held at the Hotel Radisson Blu, St. Julians, in Malta, from 17 to 22 of March, 2024. The conference will also feature the possibility to participate virtually. 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. 
#### [Conference Handbook](https://drive.google.com/file/d/13Wn38q0yev6U3RTgnf4Tn5T9GuAuZ-xW/view?usp=sharing) *downloadable*

To access the EACL 2024 event page you need to register [**here**](https://acl.swoogo.com/EACL2024). 

Please register!

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.

Recent work in open-domain question answering (ODQA) has shown that adversarial poisoning of the search collection can cause large drops in accuracy for production systems. However, little to no work has proposed methods to defend against these attacks. To do so, we rely on the intuition that redundant information often exists in large corpora. To find it, we introduce a method that uses query augmentation to search for a diverse set of passages that could answer the original question but are less likely to have been poisoned. We integrate these new passages into the model through the design of a novel confidence method, comparing the predicted answer to its appearance in the retrieved contexts (what we call Confidence from Answer Redundancy, i.e. CAR). Together these methods allow for a simple but effective way to defend against poisoning attacks that provides gains of nearly 20% exact match across varying levels of data poisoning/knowledge conflicts.

Defending Against Disinformation Attacks in Open-Domain Question Answering

In this paper, we focus on task-specific question answering (QA). To this end, we introduce a method for generating exhaustive and high-quality training data, which allows us to train compact (e.g., run on a mobile device), task-specific QA models that are competitive against GPT variants. The key technological enabler is a novel mechanism for automatic question-answer generation from procedural text which can ingest large amounts of textual instructions and produce exhaustive in-domain QA training data. While current QA data generation methods can produce well-formed and varied data, their non-exhaustive nature is sub-optimal for training a QA model. In contrast, we leverage the highly structured aspect of procedural text and represent each step and the overall flow of the procedure as graphs. We then condition on graph nodes to automatically generate QA pairs in an exhaustive and controllable manner. Comprehensive evaluations of our method show that: 1) small models trained with our data achieve excellent performance on the target QA task, even exceeding that of GPT3 and ChatGPT despite being several orders of magnitude smaller. 2) semantic coverage is the key indicator for downstream QA performance. Crucially, while large language models excel at syntactic diversity, this does not necessarily result in improvements on the end QA model. In contrast, the higher semantic coverage provided by our method is critical for QA performance.

Graph Guided Question Answer Generation for Procedural Question-Answering

Internet memes have gained significant influence in communicating political, psychological, and sociocultural ideas. While meme are often humorous, there has been a rise in the use of memes for trolling and cyberbullying. Although a wide variety of effective deep learning-based models have been developed for detecting offensive multimodal memes, only a few works have been done on explainability aspect. Recent laws like "right to explanations" of General Data Protection Regulation, have spurred research in developing interpretable models rather than only focusing on performance. Motivated by this, we introduce MultiBully-Ex, the first benchmark dataset for multimodal explanation from code-mixed cyberbullying memes. Here, both visual and textual modalities are highlighted to explain why a given meme is cyberbullying. A Contrastive Language-Image Pretraining (CLIP) projection based multimodal shared-private multitask approach has been proposed for visual and textual explanation of a meme. Experimental results demonstrate that training with multimodal explanations improves performance in generating textual justifications and more accurately identifying the visual evidence supporting a decision with reliable performance improvements.

Meme-ingful Analysis: Enhanced Understanding of Cyberbullying in Memes Through Multimodal Explanations

Large language models (LLMs) have demonstrated remarkable capability to generate fluent responses to a wide variety of user queries. However, this has also raised concerns about the potential misuse of such texts in journalism, education, and academia. In this study, we strive to create automated systems that can detect machine-generated texts and pinpoint potential misuse. We first introduce a large-scale benchmark M4, which is a multi-generator, multi-domain, and multi-lingual corpus for machine-generated text detection. Through an extensive empirical study of this dataset, we show that it is challenging for detectors to generalize well on instances from unseen domains or LLMs. In such cases, detectors tend to misclassify machine-generated text as human-written. These results show that the problem is far from solved and that there is a lot of room for improvement. We believe that our dataset will enable future research towards more robust approaches to this pressing societal problem. The dataset is available at https://github.com/mbzuai-nlp/M4

M4: Multi-generator, Multi-domain, and Multi-lingual Black-Box Machine-Generated Text Detection

Cross-lingual transfer (XLT) driven by massively multilingual language models (mmLMs) has been shown largely ineffective for low-resource (LR) target languages with little (or no) representation in mmLM's pretraining, especially if they are linguistically distant from the high-resource (HR) source language. Much of the recent focus in XLT research has been dedicated to \textit{LR language families}, i.e., families without any HR languages (e.g., families of African languages or indigenous languages of the Americas). In this work, in contrast, we investigate a configuration that is arguably of practical relevance for more of the world's languages: XLT to LR languages that do have a close HR relative. To explore the extent to which a HR language can facilitate transfer to its LR relatives, we (1) introduce Karde\c{s}-NLU, an evaluation benchmark with language understanding datasets in five LR Turkic languages: Azerbaijani, Kazakh, Kyrgyz, Uzbek, and Uyghur; and (2) investigate (a) intermediate training and (b) fine-tuning strategies that leverage Turkish in XLT to these target languages. Our experimental results show that both - integrating Turkish in intermediate training and in downstream fine-tuning - yield substantial improvements in XLT to LR Turkic languages. Finally, we benchmark cutting-edge instruction-tuned large language models on Karde\c{s}-NLU, showing that their performance is highly task- and language-dependent.

Kardeş-NLU: Transfer to Low-Resource Languages with the Help of a High-Resource Cousin – A Benchmark and Evaluation for Turkic Languages

Popular benchmarks (e.g., XNLI) used to evaluate cross-lingual language understanding consist of parallel versions of English evaluation sets in multiple target languages created with the help of professional translators. When creating such parallel data, it is critical to ensure high-quality translations for all target languages for an accurate characterization of cross-lingual transfer. In this work, we find that translation inconsistencies \emph{do exist} and interestingly they \emph{disproportionally impact low-resource languages} in XNLI. To identify such inconsistencies, we propose measuring the gap in performance between zero-shot evaluations on the human-translated and machine-translated target text across multiple target languages; relatively large gaps are indicative of translation errors. We also corroborate that translation errors exist for two target languages, namely Hindi and Urdu, by doing a manual reannotation of human-translated test instances in these two languages and finding poor agreement with the original English labels these instances were supposed to inherit.

Translation Errors Significantly Impact Low-Resource Languages in Cross-Lingual Learning

Generating natural language text from graph-structured data is essential for conversational information seeking. Semantic triples derived from knowledge graphs can serve as a valuable source for grounding responses from conversational agents by providing a factual basis for the information they communicate. This is especially relevant in the context of large language models, which offer great potential for conversational interaction but are prone to hallucinating, omitting, or producing conflicting information. In this study, we conduct an empirical analysis of conversational large language models in generating natural language text from semantic triples. We compare four large language models of varying sizes with different prompting techniques. Through a series of benchmark experiments on the WebNLG dataset, we analyze the models' performance and identify the most common issues in the generated predictions. Our findings show that the capabilities of large language models in triple verbalization can be significantly improved through few-shot prompting, post-processing, and efficient fine-tuning techniques, particularly for smaller models that exhibit lower zero-shot performance.

A Comparative Analysis of Conversational Large Language Models in Knowledge-Based Text Generation

Neural networks are notoriously data-hungry. This represents an issue in cases where data are scarce such as in low-resource languages. Data augmentation is a technique commonly used in computer vision to provide neural networks with more data and increase their generalization power. When dealing with data augmentation for natural language, however, simple data augmentation techniques similar to the ones used in computer vision such as rotation and cropping cannot be employed because they would generate ungrammatical texts. 
Thus, data augmentation needs a specific design in the case of neural logic-to-text systems, especially for a structurally rich input format such as the ones used for meaning representation. This is the case of the neural natural language generation for Discourse Representation Structures (DRS-to-Text), where the logical nature of DRS needs a specific design of data augmentation. In this paper, we adopt a novel approach in DRS-to-Text to selectively augment a training set with new data by adding and varying two specific lexical categories, i.e. proper and common nouns. In particular, we propose using WordNet supersenses to produce new training sentences using both in-and-out-of-context nouns. We present a number of experiments for evaluating the role played by augmented lexical information. The experimental results prove the effectiveness of our approach for data augmentation in DRS-to-Text generation.

Exploring Data Augmentation in Neural DRS-to-Text Generation

Despite the impressive performance of large language models (LLMs), they
often lag behind specialized models in various tasks. LLMs only use a fraction
of the existing training data for in-context learning, while task-specific
models harness the full dataset for fine-tuning. In this work, we tackle the
problem of leveraging training data to improve the performance of LLMs without
fine-tuning. Our approach directly targets LLM predictions without requiring
access to their weights. We create a pool of candidates from the LLM through
few-shot prompting and we employ a compact model, the LM-corrector (LMCor),
specifically trained to merge these candidates to produce an enhanced output.
Our experiments on four natural language generation tasks demonstrate that even
a small LMCor model (250M) substantially improves the few-shot performance of
LLMs (62B), matching and even outperforming standard fine-tuning. Furthermore,
we illustrate the robustness of LMCor against different prompts, thereby
minimizing the need for extensive prompt engineering. Finally, we show that
LMCor can be seamlessly integrated with different LLMs at inference, serving as
a plug-and-play module to improve their performance.

Small Language Models Improve Giants by Rewriting Their Outputs

Large pre-trained language models have recently been expanded and applied to programming language tasks with great success, often through further pre-training of a strictly-natural language model--where training sequences typically contain both natural and (linearised) programming language. Such approaches effectively map both modalities of the sequence into the same embedding space. However, programming language keywords (e.g. ``while'') often have very strictly defined semantics. As such, transfer learning from their natural language usage may not necessarily be beneficial to their code application and vise versa. Assuming an already pre-trained language model, in this work we investigate how sequence tokens can be adapted and represented differently, depending on which modality they belong to, and to the ultimate benefit of the downstream task. We experiment with separating embedding spaces between modalities during further model pre-training with modality-relative training objectives. We focus on text-to-code generation and observe consistent improvements across two backbone models and two test sets, measuring pass@$k$ and a novel incremental variation.

Downloads

Next from EACL 2024

Defending Against Disinformation Attacks in Open-Domain Question Answering

Similar lecture

Reducing Spurious Correlations for Answer Selection by Feature Decorrelation and Language Debiasing

.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 EACL 2024

Defending Against Disinformation Attacks in Open-Domain Question Answering

Similar lecture

Reducing Spurious Correlations for Answer Selection by Feature Decorrelation and Language Debiasing

Downloads