Thailand

This paper introduces the ColorSwap dataset, designed to assess and improve the proficiency of multimodal models in matching objects with their colors. The dataset is comprised of 2,000 unique image-caption pairs, grouped into 1,000 examples. Each example includes a caption-image pair, along with a &quot;color-swapped&quot; pair. We follow the Winoground schema: the two captions in an example have the same words, but the color words have been rearranged to modify different objects. The dataset was created through a novel blend of automated caption and image generation with humans in the loop. We evaluate image-text matching (ITM) and visual language models (VLMs) and find that even the latest ones are still not robust at this task. GPT-4V and LLaVA score 72% and 42% on our main VLM metric, although they may improve with more advanced prompting techniques. On the main ITM metric, contrastive models such as CLIP and SigLIP perform close to chance (at 12% and 30%, respectively), although the non-contrastive BLIP ITM model is stronger (87%). We also find that finetuning on fewer than 2,000 examples yields significant performance gains on this out-of-distribution word-order understanding task.

ACL 2024

ColorSwap: A Color and Word Order Dataset for Multimodal Evaluation

ai art

multimodal

benchmark

This paper introduces the ColorSwap dataset, designed to assess and improve the proficiency of multimodal models in matching objects with their colors. The dataset is comprised of 2,000 unique image-caption pairs, grouped into 1,000 examples. Each example includes a caption-image pair, along with a "color-swapped" pair. We follow the Winoground schema: the two captions in an example have the same words, but the color words have been rearranged to modify different objects. The dataset was created through a novel blend of automated caption and image generation with humans in the loop. We evaluate image-text matching (ITM) and visual language models (VLMs) and find that even the latest ones are still not robust at this task. GPT-4V and LLaVA score 72% and 42% on our main VLM metric, although they may improve with more advanced prompting techniques. On the main ITM metric, contrastive models such as CLIP and SigLIP perform close to chance (at 12% and 30%, respectively), although the non-contrastive BLIP ITM model is stronger (87%). We also find that finetuning on fewer than 2,000 examples yields significant performance gains on this out-of-distribution word-order understanding task.

poster

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

Since the natural language processing (NLP) community started to make large language models (LLMs) act as a critic to evaluate the quality of generated texts, most of the existing works train a critique generation model on the evaluation data labeled by GPT-4's direct prompting. We observe that these models lack the ability to generate informative critiques in both pointwise grading and pairwise comparison especially without references. As a result, their generated critiques cannot provide fine-grained distinguishability on generated texts, causing unsatisfactory evaluation performance. In this paper, we propose a simple yet effective method called Eval-Instruct, which can first acquire pointwise grading critiques with pseudo references and then revise these critiques via multi-path prompting to obtain informative evaluation data in different tasks and settings, including pointwise grading and pairwise comparison with / without references. After fine-tuning on these data, the resulting model CritiqueLLM is empirically shown to outperform ChatGPT and all the open-source baselines and even achieve comparable evaluation performance to GPT-4 in system-level correlations of pointwise grading. We also demonstrate that our generated critiques can act as scalable feedback to further improve the generation quality of strong LLMs like ChatGPT.

CritiqueLLM: Towards an Informative Critique Generation Model for Evaluation of Large Language Model Generation

Instruction tuning enhances the instruction following ability of large language models by finetuning with supervised instruction data. Previous work proposes in-context instruction tuning (ICIT) where specific positive or negative examples are incorporated into the prompt for better performance. In this work, we propose PACIT, a simple and effective in-context instruction tuning method, inspired by the pedagogical concept of desirable difficulty. The PACIT method unlocks the power of examples by encouraging the model to actively learn to grasp the distinctions between the positive and negative examples instead of merely reading. The model is expected to first verify the correctness of the provided example according to the task description, which is then set as the condition for generating a better response to the task instance. Our extensive experiments prove the effectiveness of PACIT, outperforming ICIT baseline on both in-domain and out-domain tasks up to 9.16 and 3.14 average ROUGE-L scores, respectively. Moreover, PACIT can notably enhance the performance of instruction tuning even when all positive and negative examples are generated with a self-instruct method.

PACIT: Unlocking the Power of Examples for Better In-Context Instruction Tuning

Deploying large language models (LLMs) to real scenarios for domain-specific question answering (QA) is a key thrust for LLM applications, which poses numerous challenges, especially in ensuring that responses are both accommodating to user requirements and appropriately leveraging domain-specific knowledge bases. They are the two major difficulties for LLM application as vanilla fine-tuning falls short of addressing. Combining these requirements, we conceive of them as the requirement for the model's preference to be harmoniously aligned with humans'. Thus, we introduce Knowledgeable Preference AlignmenT (KnowPAT), which constructs two kinds of preference sets to tackle the two issues. Besides, we design a new alignment objective to align the LLM preference with different human preferences uniformly, aiming to optimize LLM performance in real-world, domain-specific QA settings. Adequate experiments and comprehensive comparisons with 15 baseline methods illustrate that our KnowPAT is a superior pipeline for real-scenario domain-specific QA with LLMs.

Knowledgeable Preference Alignment for LLMs in Domain-specific Question Answering

Neural Machine Translation (NMT) continues to improve in quality and adoption, yet the in advertent perpetuation of gender bias remains a significant concern. Despite numerous studies on gender bias in translations into English from weakly gendered-languages, there are no benchmarks for evaluating this phenomenon or for assessing mitigation strategies. To address this gap, we introduce GATE X-E, an extension to the GATE (Rarrick et al., 2023) corpus, that consists of human translations from Turkish, Hungarian, Finnish, and Persian into English. Each translation is accompanied by feminine, masculine, and neutral variants. The dataset, which contains between 1250 and 1850 instances for each of the four language pairs, features natural sentences with a wide range of sentence lengths and domains, challenging translation rewriters on various linguistic phenomena. Additionally, we present a translation gender rewriting solution built with GPT-4 and use GATE X-E to evaluate it. We open source our contributions to encourage further research on gender debiasing.

GATE X-E : A Challenge Set for Gender-Fair Translations from Weakly-Gendered Languages

To date, toxicity mitigation in language models has almost entirely been focused on single-language settings. As language models embrace multilingual capabilities, it's crucial our safety measures keep pace. Recognizing this research gap, our approach expands the scope of conventional toxicity mitigation to address the complexities presented by multiple languages. In the absence of sufficient annotated datasets across languages, we employ translated data to evaluate and enhance our mitigation techniques. We also compare finetuning mitigation approaches against retrieval-augmented techniques under both static and continual toxicity mitigation scenarios. This allows us to examine the effects of translation quality and the cross-lingual transfer on toxicity mitigation. We also explore how model size and data quantity affect the success of these mitigation efforts. Covering nine languages, our study represents a broad array of linguistic families and levels of resource availability, ranging from high to mid-resource languages. Through comprehensive experiments, we provide insights into the complexities of multilingual toxicity mitigation, offering valuable insights and paving the way for future research in this increasingly important field.

From One to Many: Expanding the Scope of Toxicity Mitigation in Language Models

Visual Grounding (VG) methods in Visual Question Answering (VQA) attempt to improve VQA performance by strengthening a model's reliance on question-relevant visual information. The presence of such relevant information in the visual input is typically assumed in training and testing. This assumption, however, is inherently flawed when dealing with imperfect image representations common in large-scale VQA, where the information carried by visual features frequently deviates from expected ground-truth contents. As a result, training and testing of VG-methods is performed with largely inaccurate data, which obstructs proper assessment of their potential benefits.

In this study, we demonstrate that current evaluation schemes for VG-methods are problematic due to the flawed assumption of availability of relevant visual information. Our experiments show that these methods can be much more effective when evaluation conditions are corrected. Code is provided.

Uncovering the Full Potential of Visual Grounding Methods in VQA

Vision-and-language (VL) models with separate encoders for each modality (e.g., CLIP) have become the go-to models for zero-shot image classification and image-text retrieval. They are, however, mostly evaluated in English as multilingual benchmarks are limited in availability. We introduce Babel-ImageNet, a massively multilingual benchmark that offers (partial) translations of ImageNet labels to 100 languages, built without machine translation or manual annotation. We instead automatically obtain reliable translations by linking them -- via shared WordNet synsets -- to BabelNet, a massively multilingual lexico-semantic network. We evaluate 11 public multilingual CLIP models on zero-shot image classification (ZS-IC) on our benchmark, demonstrating a significant gap between English ImageNet performance and that of high-resource languages (e.g., German or Chinese), and an even bigger gap for low-resource languages (e.g., Sinhala or Lao). Crucially, we show that the models' ZS-IC performance highly correlates with their performance in image-text retrieval, validating the use of Babel-imageNet to evaluate multilingual models for the vast majority of languages without gold image-text data. Finally, we show that the performance of multilingual CLIP can be drastically improved for low-resource languages with parameter-efficient language-specific training. 
We make our code and data publicly available: \url{https://github.com/gregor-ge/Babel-ImageNet}

Babel-ImageNet: Massively Multilingual Evaluation of Vision-and-Language Representations

We analyze the operation of transformer language adapters, which are small modules trained on top of a frozen language model to adapt its predictions to new target languages. We show that adapted predictions mostly evolve in the source language the model was trained on, while the target language becomes pronounced only in the very last layers of the model. Moreover, the adaptation process is gradual and distributed across layers, where it is possible to skip small groups of adapters without decreasing adaptation performance. Last, we show that adapters operate on top of the model’s frozen representation space while largely preserving its structure, rather than on an isolated subspace. Our findings provide a deeper view into the adaptation process of language models to new languages, showcasing the constraints imposed on it by the underlying model and introduces practical implications to enhance its efficiency.

The Hidden Space of Transformer Language Adapters

Impressive performance of pre-trained models has garnered public attention and made news headlines in recent years. Almost always, these models are produced by or in collaboration with industry. Using them is critical for competing on natural language processing (NLP) benchmarks and correspondingly to stay relevant in NLP research. We surveyed 100 papers published at EMNLP 2022 to determine the degree to which researchers rely on industry models, other artifacts, and contributions to publish in prestigious NLP venues and found that the ratio of their citation is at least three times greater than what would be expected. Our work serves as a scaffold to enable future researchers to more accurately address whether: 1) Collaboration with industry is still collaboration in the absence of an alternative or 2) if NLP inquiry has been captured by the motivations and research direction of private corporations.

Collaboration or Corporate Capture? Quantifying NLP's Reliance on Industry Artifacts and Contributions

Large language models demonstrate a remarkable capability for learning to solve new tasks from a few examples.
The $\textit{prompt template}$, or the way the input examples are formatted to obtain the prompt, is an important yet often overlooked aspect of in-context learning.
In this work, we conduct a comprehensive study of the template format's influence on the in-context learning performance.
We evaluate the impact of the prompt template across 21 models (from 770M to 70B parameters) and 4 standard classification datasets. 
We show that a poor choice of the template can reduce the performance of the strongest models and inference methods to a random guess level.
More importantly, the best templates do not transfer between different setups and even between models of the same family.
Our findings show that the currently prevalent approach to evaluation, which ignores template selection, may give misleading results due to different templates in different works.
As a first step towards mitigating this issue, we propose $\textit{Template Ensembles}$ that aggregate model predictions across several templates.
This simple test-time augmentation boosts average performance while being robust to the choice of random set of templates.

Premium content

ColorSwap: A Color and Word Order Dataset for Multimodal Evaluation

Downloads

Next from ACL 2024

CritiqueLLM: Towards an Informative Critique Generation Model for Evaluation of Large Language Model Generation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES