Social Learning: Towards Collaborative Learning with Large Language Models

We introduce the framework of "social learning" in the context of large language models (LLMs), whereby models share knowledge with each other in a privacy-aware manner using natural language. We present and evaluate two approaches for knowledge transfer between LLMs. In the first scenario, we allow the model to generate abstract prompts aiming to teach the task. In our second approach, models transfer knowledge by generating synthetic examples. We evaluate these methods across diverse datasets and quantify memorization as a proxy for privacy loss. These techniques inspired by social learning yield promising results with low memorization of the original data. In particular, we show that performance using these methods is comparable to results with the use of original labels and prompts. Our work demonstrates the viability of social learning for LLMs, establishes baseline approaches and highlights several unexplored areas for future work.

The Impact of Preference Agreement in Reinforcement Learning from Human Feedback: A Case Study in Summarization

Reinforcement Learning from Human Feedback (RLHF) can be used to capture complex and nuanced properties of text generation quality. As a result, the task of text summarization has been identified as a good candidate for this process. In this paper, we explore how preference agreement impacts the efficacy of RLHF for summarization. We show that sampling human preferences to include a range of annotator agreement results in (1) higher accuracy reward models and (2) alters the characteristics of quality captured. We additionally show improvements in downstream generation when using a reward model trained with a range of preference agreements. Our contributions have implications for the design of synthetic datasets as well as the importance of considering quality differentials in comparison-based data.

Towards Better Evaluation of Instruction-Following: A Case-Study in Summarization

Despite recent advances, evaluating how well large language models (LLMs) follow user instructions remains an open problem. While evaluation methods of language models have seen a rise in prompt-based approaches, limited work on the correctness of these methods has been conducted. In this work, we perform a meta-evaluation of a variety of metrics to quantify how accurately they measure the instruction-following abilities of LLMs. Our investigation is performed on grounded query-based summarization by collecting a new short-form, real-world dataset riSum, containing 300 document-instruction pairs with 3 answers each. All 900 answers are rated by 3 human annotators. Using riSum, we analyze the agreement between evaluation methods and human judgment. Finally, we propose new LLM-based reference-free evaluation methods that improve upon established baselines and perform on par with costly reference-based metrics that require high-quality summaries.

One Size Does Not Fit All: The Case for Personalised Word Complexity Models

Complex Word Identification (CWI) aims to detect words within a text that a reader may find difficult to understand. It has been shown that CWI systems can improve text simplification, readability prediction and vocabulary acquisition modelling. However, the difficulty of a word is a highly idiosyncratic notion that depends on a reader's first language, proficiency and reading experience. In this paper, we show that personal models are best when predicting word complexity for individual readers. We use a novel active learning framework that allows models to be tailored to individuals and release a dataset of complexity annotations and models as a benchmark for further research.

On the Ethical Considerations of Text Simplification

This paper outlines the ethical implications of text simplification within the framework of assistive systems. We argue that a distinction should be made between the technologies that perform text simplification and the realisation of these in assistive technologies. When using the latter as a motivation for research, it is important that the subsequent ethical implications be carefully considered. We provide guidelines for the framing of text simplification independently of assistive systems, as well as suggesting directions for future research and discussion based on the concerns raised.

Predicting Text Readability from Scrolling Interactions

Judging the readability of text has many important applications, for instance when performing text simplification or when sourcing reading material for language learners. In this paper, we present a 518 participant study which investigates how scrolling behaviour relates to the readability of a text. We make our dataset publicly available and show that (1) there are statistically significant differences in the way readers interact with text depending on the text level, (2) such measures can be used to predict the readability of text, and (3) the background of a reader impacts their reading interactions and the factors contributing to text difficulty.

Detecting Multiword Expression Type Helps Lexical Complexity Assessment

Multiword expressions (MWEs) represent lexemes that should be treated as single lexical units due to their idiosyncratic nature. Multiple NLP applications have been shown to benefit from MWE identification, however the research on lexical complexity of MWEs is still an under-explored area. In this work, we re-annotate the Complex Word Identification Shared Task 2018 dataset of Yimam et al. (2017), which provides complexity scores for a range of lexemes, with the types of MWEs. We release the MWE-annotated dataset with this paper, and we believe this dataset represents a valuable resource for the text simplification community. In addition, we investigate which types of expressions are most problematic for native and non-native readers. Finally, we show that a lexical complexity assessment system benefits from the information about MWE types.