Thailand

Large language models (LLMs) can explain their predictions through post-hoc or Chain-of-Thought (CoT) explanations. But an LLM could make up reasonably sounding explanations that are unfaithful to its underlying reasoning. Recent work has designed tests that aim to judge the faithfulness of post-hoc or CoT explanations. In this work we argue that these faithfulness tests do not measure faithfulness to the models&#39; inner workings -- but rather their self-consistency at output level.
Our contributions are three-fold: i) We clarify the status of faithfulness tests in view of model explainability, characterising them as self-consistency tests instead. This assessment we underline by ii) constructing a Comparative Consistency Bank for self-consistency tests that for the first time compares existing tests on a common suite of 11 open LLMs and 5 tasks -- including iii) our new self-consistency measure CC-SHAP. CC-SHAP is a fine-grained measure (not a test) of LLM self-consistency. It compares how a model&#39;s input contributes to the predicted answer and to generating the explanation. Our fine-grained CC-SHAP metric allows us iii) to compare LLM behaviour when making predictions and to analyse the effect of other consistency tests at a deeper level, which takes us one step further towards measuring faithfulness by bringing us closer to the internals of the model than strictly surface output-oriented tests.

ACL 2024

On Measuring Faithfulness or Self-consistency of Natural Language Explanations

self-consistency

natural language explanations

shap

faithfulness

interpretability

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

EMNLP 2025

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Large language models (LLMs) can explain their predictions through post-hoc or Chain-of-Thought (CoT) explanations. But an LLM could make up reasonably sounding explanations that are unfaithful to its underlying reasoning. Recent work has designed tests that aim to judge the faithfulness of post-hoc or CoT explanations. In this work we argue that these faithfulness tests do not measure faithfulness to the models' inner workings -- but rather their self-consistency at output level.
Our contributions are three-fold: i) We clarify the status of faithfulness tests in view of model explainability, characterising them as self-consistency tests instead. This assessment we underline by ii) constructing a Comparative Consistency Bank for self-consistency tests that for the first time compares existing tests on a common suite of 11 open LLMs and 5 tasks -- including iii) our new self-consistency measure CC-SHAP. CC-SHAP is a fine-grained measure (not a test) of LLM self-consistency. It compares how a model's input contributes to the predicted answer and to generating the explanation. Our fine-grained CC-SHAP metric allows us iii) to compare LLM behaviour when making predictions and to analyse the effect of other consistency tests at a deeper level, which takes us one step further towards measuring faithfulness by bringing us closer to the internals of the model than strictly surface output-oriented tests.

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.

Instruction Fine-tuning (IFT) is a crucial phase in building large language models (LLMs). Previous works mainly focus on the IFT's role in the transfer of behavioral norms and the learning of additional world knowledge. However, the understanding of the underlying mechanisms of IFT remains significantly limited. In this paper, we design a knowledge intervention framework to decouple the potential underlying factors of IFT, thereby enabling individual analysis of different factors. Surprisingly, our experiments reveal that attempting to learn additional world knowledge through IFT often struggles to yield positive impacts and can even lead to markedly negative effects. Further, we discover that maintaining internal knowledge consistency before and after IFT is a critical factor for achieving successful IFT. Our findings reveal the underlying mechanisms of IFT and provide robust support for some very recent and potential future works.

Learning or Self-aligning? Rethinking Instruction Fine-tuning

Recent progress in LLMs discussion suggests that multi-agent discussion improves the reasoning abilities of LLMs. In this work, we reevaluate this claim through systematic experiments, where we propose a novel group discussion framework to enrich the set of discussion mechanisms. Interestingly, our results show that a single-agent LLM with strong prompts can achieve almost the same best performance as the best existing discussion approach on a wide range of reasoning tasks and backbone LLMs. We observed that the multi-agent discussion performs better than a single agent only when there is no demonstration in the prompt. Further study reveals the common interaction mechanisms of LLMs during the discussion. Our code can be found in \url{https://github.com/HKUST-KnowComp/LLM-discussion}.

Rethinking the Bounds of LLM Reasoning: Are Multi-Agent Discussions the Key?

While auxiliary information has become a key to enhancing Large Language Models (LLMs), relatively little is known about how LLMs merge these contexts, specifically contexts generated by LLMs and those retrieved from external sources.
To investigate this, we formulate a systematic framework to identify whether LLMs' responses are attributed to either generated or retrieved contexts.
To easily trace the origin of the response, we construct datasets with conflicting contexts, i.e., each question is paired with both generated and retrieved contexts, yet only one of them contains the correct answer.
Our experiments reveal a significant bias in several LLMs (GPT-4/3.5 and Llama2) to favor generated contexts, even when they provide incorrect information.
We further identify two key factors contributing to this bias: i) contexts generated by LLMs typically show greater similarity to the questions, increasing their likelihood of being selected; ii) the segmentation process used in retrieved contexts disrupts their completeness, thereby hindering their full utilization in LLMs.
Our analysis enhances the understanding of how LLMs merge diverse contexts, offers valuable insights for advancing current LLM augmentation methods, and highlights the risk of generated misinformation for retrieval-augmented LLMs.

Blinded by Generated Contexts: How Language Models Merge Generated and Retrieved Contexts When Knowledge Conflicts?

Identifying the training datasets that influence a language model's outputs is essential for minimizing the generation of harmful content and enhancing its performance. Ideally, we can measure the influence of each dataset by removing it from training; however, it is prohibitively expensive to retrain a model multiple times. This paper presents UnTrac: unlearning traces the influence of a training dataset on the model's performance. UnTrac is extremely simple; each training dataset is unlearned by gradient ascent, and we evaluate how much the model's predictions change after unlearning. Furthermore, we propose a more scalable approach, UnTrac-Inv, which unlearns a test dataset and evaluates the unlearned model on training datasets. UnTrac-Inv resembles UnTrac, while being efficient for massive training datasets. In the experiments, we examine if our methods can assess the influence of pretraining datasets on generating toxic, biased, and untruthful content. Our methods estimate their influence much more accurately than existing methods while requiring neither excessive memory space nor multiple checkpoints.

Unlearning Traces the Influential Training Data of Language Models

Supervised approaches to probing large language models (LLMs) have been criticized of using pre-defined theory-laden target labels. As an alternative, parameter-free probing constructs structural representations bottom-up via information derived from the LLM alone. This has been suggested to capture a genuine “LLM-internal grammar”. However, its relation to familiar linguistic formalisms remains unclear. I extend prior work on a parameter-free probing technique called perturbed masking applied to BERT, by comparing its results to the Universal Dependencies (UD) formalism for English. The results highlight several major discrepancies between BERT and UD, which lack correlates in linguistic theory. This raises the question of whether human grammar is the correct analogy to interpret BERT in the first place.

What Does Parameter-free Probing Really Uncover?

Despite their remarkable ability to capture linguistic nuances across diverse languages, questions persist regarding the degree of alignment between languages in multilingual embeddings. Drawing inspiration from research on high-dimensional representations in neural language models, we employ clustering to uncover latent concepts within multilingual models. Our analysis focuses on quantifying the alignment and overlap of these concepts across various languages within the latent space. To this end, we introduce two metrics CALIGN and COLAP aimed at quantifying these aspects, enabling a deeper exploration of multilingual embeddings. Our study encompasses three multilingual models (mT5, mBERT, and XLM-R) and three downstream tasks (Machine Translation, Named Entity Recognition, and Sentiment Analysis). Key findings from our analysis include: i) deeper layers in the network demonstrate increased cross-lingual alignment due to the presence of language-agnostic concepts, ii) fine-tuning of the models enhances alignment within the latent space, and iii) such task-specific calibration helps in explaining the emergence of zero-shot capabilities in the models.

Exploring Alignment in Shared Cross-lingual Spaces

Multi-modal large language models (MLLMs) have demonstrated remarkable performance across various tasks. However, these models often prioritize capturing global information and overlook the importance of perceiving local information. This limitation hinders their ability to effectively understand fine-grained details and handle grounding tasks that necessitate nuanced comprehension. Although some recent works have made strides in this, they have primarily focused on single-modality inputs. Therefore, we propose \textbf{GroundingGPT}, an end-to-end language enhanced multi-modal grounding model. It is designed to perform fine-grained grounding tasks for three modalities: image, video and audio. To enhance the model's performance, we adopt a coarse-to-fine training strategy, utilizing a three-stage training approach to progressively enhance the model's semantic awareness and fine-grained understanding capabilities. Additionally, we employ a diversified stage-specific dataset construction pipeline, developing a multi-modal, multi-granularity dataset tailored for training the model in different stages. Extensive experiments conducted on multiple multi-modal benchmarks demonstrate that our model achieves impressive fine-grained understanding of multi-modal inputs on grounding tasks while maintaining or improving its global comprehension capabilities. Our code, model, and dataset are available at https://github.com/lzw-lzw/GroundingGPT.

GroundingGPT: Language Enhanced Multi-modal Grounding Model

Large language models (LLMs) have an impressive ability to draw on novel information supplied in their context. Yet the mechanisms underlying this contextual grounding remain unknown, especially in situations where contextual information contradicts factual knowledge stored in the parameters, which LLMs also excel at recalling. Favoring the contextual information is critical for retrieval-augmented generation methods, which enrich the context with up-to-date information, hoping that grounding can rectify outdated or noisy stored knowledge. We present a novel method to study grounding abilities using Fakepedia, a novel dataset of counterfactual texts constructed to clash with a model's internal parametric knowledge. In this study, we introduce Fakepedia, a counterfactual dataset designed to evaluate grounding abilities when the internal parametric knowledge clashes with the contextual information. We benchmark various LLMs with Fakepedia and conduct a causal mediation analysis of LLM components when answering Fakepedia queries, based on our Masked Grouped Causal Tracing (MGCT) method. Through this analysis, we identify distinct computational patterns between grounded and ungrounded responses. We finally demonstrate that distinguishing grounded from ungrounded responses is achievable through computational analysis alone. Our results, together with existing findings about factual recall mechanisms, provide a coherent narrative of how grounding and factual recall mechanisms interact within LLMs.

A Glitch in the Matrix? Locating and Detecting Language Model Grounding with Fakepedia

Multipanel images, commonly seen as web screenshots, posters, etc., pervade our daily lives. These images, characterized by their composition of multiple subfigures in distinct layouts, effectively convey information to people. Toward building advanced multimodal AI applications, such as agents that understand complex scenes and navigate through webpages, the skill of multipanel visual reasoning is essential, and a comprehensive evaluation of models in this regard is important. Therefore, we introduce Multipanel Visual Question Answering (MultipanelVQA), a novel benchmark comprising 6,600 triplets of questions, answers, and multipanel images that specifically challenge models in comprehending multipanel images. Our evaluation shows that questions in the MultipanelVQA benchmark pose significant challenges to the state-of-the-art Multimodal Large Language Models (MLLMs) tested, even though humans can attain approximately 99\% accuracy on these questions. Distinctively, the MultipanelVQA benchmark features synthetically generated multipanel images specifically crafted to isolate and assess the impact of various factors, such as the layout, on MLLMs' multipanel image comprehension abilities. As a result, in addition to benchmarking the capabilities of MLLMs in understanding multipanel images, we analyze various factors of the multipanel image that affect MLLMs' performance with synthetic data and offer insights for enhancement.

Muffin or Chihuahua? Challenging Multimodal Large Language Models with Multipanel VQA

This study investigates the existence of positional biases in Transformer-based language models for text representation learning, particularly in the context of web document retrieval. We build on previous research that demonstrated loss of information in the middle of input sequences for causal language models, extending it to the domain of embedding learning. We examine positional biases at multiple stages of the training pipeline for an encoder-decoder neural retrieval model, namely language model pre-training, contrastive pre-training, and contrastive fine-tuning. Experiments with the MS-MARCO document collection reveal that after contrastive pre-training the model already generates embeddings that better capture the beginning of the input content, with fine-tuning further aggravating this effect.

Downloads

Next from ACL 2024

Learning or Self-aligning? Rethinking Instruction Fine-tuning

Similar lecture

Towards Faithful Natural Language Explanations: A Study Using Activation Patching in Large Language Models