Singapore

High-risk domains pose unique challenges that require language models to provide accurate and safe responses. Despite the great success of large language models (LLMs), such as ChatGPT and its variants, their performance in high-risk domains remains unclear. Our study delves into an in-depth analysis of the performance of instruction-tuned LLMs, focusing on factual accuracy and safety adherence. To comprehensively assess the capabilities of LLMs, we conduct experiments on six NLP datasets including question answering and summarization tasks within two high-risk domains: legal and medical. Further qualitative analysis highlights the existing limitations inherent in current LLMs when evaluating in high-risk domains. This underscores the essential nature of not only improving LLM capabilities but also prioritizing the refinement of domain-specific metrics, and embracing a more human-centric approach to enhance safety and factual reliability. Our findings advance the field toward the concerns of properly evaluating LLMs in high-risk domains, aiming to steer the adaptability of LLMs in fulfilling societal obligations and aligning with forthcoming regulations, such as the EU AI Act.

EMNLP 2023

Walking a Tightrope -- Evaluating Large Language Models in High-Risk Domains

high-risk domain

instruction tuning

large language models

domain adaptation

workshop paper

**Welcome to EMNLP 2023!**

On behalf of the EMNLP 2023 Organizing Committee, I extend a warm and heartfelt welcome to all of you to the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). It is with immense pleasure and excitement that we gather here in Singapore, a vibrant hub of innovation and technological advancement.

The conference program is packed with insightful presentations, thought-provoking workshops, and engaging networking opportunities. In addition to the technical sessions, we have also planned several social events that will provide you with the opportunity to connect with your colleagues.

The conference is held in person at the Resorts World Convention Centre in Singapore, and available on-line with the help of Underline.

For those in Singapore, I hope that you will find time to explore the exciting Sentosa Island, Gardens by the Bay, Singapore Botanic Gardens and the many other attractions unique to Singapore.

Of course, we are grateful to our sponsors and partners for their generous support of EMNLP 2023, whose contributions make it possible for us to host this world-class event.

Yuji Matsumoto (RIKEN AIP) 
EMNLP 2023 General Chair

To access the **EMNLP 2023** event page on Underline, you need to register for the Conference. 
Please follow **[this link](https://2023.emnlp.org/registration/)** for more details.

EMNLP 2023 took place in Singapore from Dec 6th to Dec 10th, 2023.

With the ever-growing presence of social media platforms comes the increased spread of harmful content and the need for robust hate speech detection systems.
Such systems easily overfit to specific targets and keywords, and evaluating them without considering distribution shifts that might occur between train and test data overestimates their benefit.
We challenge hate speech models via new train-test splits of existing datasets that rely on the clustering of models' hidden representations.
We present two split variants (SUBSET-SUM-SPLIT and CLOSEST-SPLIT) that, when applied to two datasets using four pretrained models, reveal how models catastrophically fail on blind spots in the latent space.
This result generalises when developing a split with one model and evaluating it on another.
Our analysis suggests that there is no clear surface-level property of the data split that correlates with the decreased performance, which underscores that task difficulty is not always humanly interpretable.
We recommend incorporating latent feature-based splits in model development and release two splits via the GenBench benchmark.

Latent Feature-based Data Splits to Improve Generalisation Evaluation: A Hate Speech Detection Case Study

Compositional generalization, the ability of intelligent models to extrapolate understanding of components to novel compositions, is a fundamental yet challenging facet in AI research, especially within multimodal environments. In this work, we address this challenge by exploiting the syntactic structure of language to boost compositional generalization. This paper elevates the importance of syntactic grounding, particularly through attention-masking techniques derived from text input parsing. We introduce and evaluate the merits of using syntactic information in the multimodal grounding problem. Our results on grounded compositional generalization underscore the positive impact of dependency parsing across diverse tasks when utilized with Weight Sharing across the Transformer encoder. The results push the state-of-the-art in multimodal grounding and parameter-efficient modeling and provide insights for future research.

Syntax-Guided Transformers: Elevating Compositional Generalization and Grounding in Multimodal Environments

Language models achieve remarkable results on a variety of tasks, yet still struggle with compositional generalisation benchmarks. The majority of these benchmarks evaluate perfor- mance in English only, leaving us with the question of whether these results generalise to other languages. As an initial step to an- swering this question, we introduce mSCAN, a multilingual adaptation of the SCAN dataset. It was produced by a rule-based translation, de- veloped in cooperation with native speakers. We then showcase this novel dataset on some in-context learning experiments, with the mul- tilingual large language model BLOOM as well as gpt3.5-turbo.

mSCAN: A Dataset for Multilingual Compositional Generalisation Evaluation | VIDEO

Due to the finite nature of any evidence used in learning, systematic generalization is crucially reliant on the presence of inductive bias (Mitchell, 1980). We examine inductive biases in different types of sequence-to-sequence neural network models, including CNNs, LSTMs (with and without attention), and transformers, inspired by Kharitonov and Chaabouni (2021). Crucially, however, we consider a wider range of possible inductive biases than their study did. Investigating preferences for hierarchical generalization compared to other types of generalization, we find that, contrary to their results, transformers display no preference for hierarchical generalization, but instead prefer a counting strategy. We also investigate biases toward different types of compositionality. By controlling for a confound in Kharitonov and Chaabouni (2021)'s test set, we find much less consistent generalization overall, and find that a large number of responses were among types other than the two types of generalization they had considered. Nevertheless, we observe consistent compositional generalization to held out combinations of primitives and functions on a SCAN task (Lake and Baroni, 2017) by models of all types, but only when primitives occur with other functions in the training set. The pattern of success indicates generalization in models of these types is highly sensitive to distributional properties of their training data.

Inductive Bias Is in the Eye of the Beholder

To develop a system with near-human language capabilities, we need to understand current systems' generalisation and compositional abilities. We approach this by generating compositional, structured data, inspired from visual intelligence tests, that depend on the problem-solvers being able to disentangle objects and their absolute and relative properties in a sequence of images. 
We design an analogous task and develop the corresponding datasets that capture specific linguistic phenomena and their properties. Solving each problem instance depends on detecting the relevant linguistic objects and generative rules of the problem. We propose two datasets modelling two linguistic phenomena -- subject-verb agreement in French, and verb alternations in English. The datasets can be used to investigate how LLMs encode linguistic objects, such as phrases, their grammatical and semantic properties, such as number or semantic role, and how such information is combined to correctly solve each problem. Specifically generated error types help investigate the behaviour of the system, which important information it is able to detect, and which structures mislead it.

Blackbird Language Matrices Tasks for Generalization

In-context learning (ICL) using large language models for tasks with many labels is challenging due to the limited context window, which makes it difficult to fit a sufficient number of examples in the prompt. In this paper, we use a pre-trained dense retrieval model to bypass this limitation, giving the model only a partial view of the full label space for each inference call. Testing with recent open-source LLMs (OPT, LLaMA), we set new state of the art performance in few-shot settings for three common intent classification datasets, with no fine-tuning. We also surpass fine-tuned performance on fine-grained sentiment classification in certain cases. We analyze the performance across number of in-context examples and different model scales, showing that larger models are necessary to effectively make use of larger context lengths for ICL. By running several ablations, we analyze the model's use of: a) the similarity of the in-context examples to the current input, b) the semantic content of the class names, and c) the correct correspondence between examples and labels. We demonstrate that all three are needed to varying degrees depending on the domain, contrary to certain recent works.

In-Context Learning for Text Classification with Many Labels

GQG is a dataset with 18,360 prompts consisting of various quantifier expressions to evaluate the generalization abilities of language models. GQG dataset enables research on different axes of generalization, such as token frequency.

GQG: Generalized Quantifier Generalization - A Dataset for Evaluating Quantifier Semantics Understanding in Language Models

This paper presents an innovative data augmentation framework with data quality control designed to enhance the robustness of Question Answering (QA) models in low-resource languages, particularly Thai. Recognizing the challenges posed by the scarcity and quality of training data, we leverage data augmentation techniques in both monolingual and cross-lingual settings. Our approach augments and enriches the original dataset, thereby increasing its linguistic diversity and robustness. We evaluate the robustness of our framework on Machine Reading Comprehension, and the experimental results illustrate the potential of data augmentation to effectively increase training data and improve model generalization in low-resource language settings, offering a promising direction for the data augmentation manner.

Cross-Lingual Data Augmentation For Thai Question-Answering | VIDEO

Compositional generalisation (CG), in NLP and in machine learning more generally, has been assessed mostly using artificial datasets. It is important to develop benchmarks to assess CG also in real-world natural language tasks in order to understand the abilities and limitations of systems deployed in the wild. To this end, our GenBench Collaborative Benchmarking Task submission utilises the distribution-based compositionality assessment (DBCA) framework to split the Europarl translation corpus into a training and a test set in such a way that the test set requires compositional generalisation capacity. Specifically, the training and test sets have divergent distributions of dependency relations, testing NMT systems' capability of translating dependencies that they have not been trained on. This is a fully-automated procedure to create natural language compositionality benchmarks, making it simple and inexpensive to apply it further to other datasets and languages. The code and data for the experiments is available at https://github.com/aalto-speech/dbca.

On using distribution-based compositionality assessment to evaluate compositional generalisation in machine translation

Semantic parsing plays a pivotal role in advancing the accessibility of human-computer interaction on a large scale. Spider, a widely recognized dataset for text-to-SQL, contains a wide range of natural language (NL) questions in English and corresponding SQL queries. 
Original splits of Spider and its adapted to Russian language and improved version, PAUQ, assume independence and identical distribution of training and testing data (i.i.d split). In this work, we propose a target length split and multilingual i.i.d split to measure compositionality and cross-language generalization. We present experimental results of popular text-to-SQL models on original, multilingual, and target length splits. We also construct a context-free grammar for the evaluation of compositionality in text-to-SQL in an out-of-distribution setting. We make the splits publicly available on HuggingFace hub via https://huggingface.co/datasets/composite/pauq.

Premium content

Walking a Tightrope -- Evaluating Large Language Models in High-Risk Domains

Downloads

Next from EMNLP 2023

Latent Feature-based Data Splits to Improve Generalisation Evaluation: A Hate Speech Detection Case Study