Why does in-context learning fail sometimes? Evaluating in-context learning on open and closed questions

• URL: http://arxiv.org/abs/2407.02028v1
• Date: Tue, 2 Jul 2024 07:52:30 GMT
• Title: Why does in-context learning fail sometimes? Evaluating in-context learning on open and closed questions
• Authors: Xiang Li, Haoran Tang, Siyu Chen, Ziwei Wang, Ryan Chen, Marcin Abram,
• Abstract summary: We measure the performance of in-context learning as a function of task novelty and difficulty for open and closed questions. We show that counter-intuitively, a context that is more aligned with the topic does not always help more than a less relevant context.
• Score: 14.999106867218572
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We measure the performance of in-context learning as a function of task novelty and difficulty for open and closed questions. For that purpose, we created a novel benchmark consisting of hard scientific questions, each paired with a context of various relevancy. We show that counter-intuitively, a context that is more aligned with the topic does not always help more than a less relevant context. This effect is especially visible for open questions and questions of high difficulty or novelty. This result reveals a fundamental difference between the treatment of close-form and open-form questions by large-language models and shows a need for a more robust evaluation of in-context learning on the variety of different types of questions. It also poses a new question of how to optimally select a context for large language models, especially in the context of Retrieval Augmented Generation (RAG) systems. Our results suggest that the answer to this question can be highly application-dependent and might be contingent on factors including the format of the question, the perceived difficulty level of the questions, and the novelty or popularity of the information we seek.

Related papers

• How to Engage Your Readers? Generating Guiding Questions to Promote Active Reading [60.19226384241482]
  We introduce GuidingQ, a dataset of 10K in-text questions from textbooks and scientific articles. We explore various approaches to generate such questions using language models. We conduct a human study to understand the implication of such questions on reading comprehension.
  arXiv  Detail & Related papers  (2024-07-19T13:42:56Z)
• Which questions should I answer? Salience Prediction of Inquisitive Questions [118.097974193544]
  We show that highly salient questions are empirically more likely to be answered in the same article. We further validate our findings by showing that answering salient questions is an indicator of summarization quality in news.
  arXiv  Detail & Related papers  (2024-04-16T21:33:05Z)
• Qsnail: A Questionnaire Dataset for Sequential Question Generation [76.616068047362]
  We present the first dataset specifically constructed for the questionnaire generation task, which comprises 13,168 human-written questionnaires. We conduct experiments on Qsnail, and the results reveal that retrieval models and traditional generative models do not fully align with the given research topic and intents. Despite enhancements through the chain-of-thought prompt and finetuning, questionnaires generated by language models still fall short of human-written questionnaires.
  arXiv  Detail & Related papers  (2024-02-22T04:14:10Z)
• Answering Ambiguous Questions with a Database of Questions, Answers, and Revisions [95.92276099234344]
  We present a new state-of-the-art for answering ambiguous questions that exploits a database of unambiguous questions generated from Wikipedia. Our method improves performance by 15% on recall measures and 10% on measures which evaluate disambiguating questions from predicted outputs.
  arXiv  Detail & Related papers  (2023-08-16T20:23:16Z)
• RECKONING: Reasoning through Dynamic Knowledge Encoding [51.076603338764706]
  We show that language models can answer questions by reasoning over knowledge provided as part of the context. In these situations, the model fails to distinguish the knowledge that is necessary to answer the question. We propose teaching the model to reason more robustly by folding the provided contextual knowledge into the model's parameters.
  arXiv  Detail & Related papers  (2023-05-10T17:54:51Z)
• Discourse Comprehension: A Question Answering Framework to Represent Sentence Connections [35.005593397252746]
  A key challenge in building and evaluating models for discourse comprehension is the lack of annotated data. This paper presents a novel paradigm that enables scalable data collection targeting the comprehension of news documents. The resulting corpus, DCQA, consists of 22,430 question-answer pairs across 607 English documents.
  arXiv  Detail & Related papers  (2021-11-01T04:50:26Z)
• Inquisitive Question Generation for High Level Text Comprehension [60.21497846332531]
  We introduce INQUISITIVE, a dataset of 19K questions that are elicited while a person is reading through a document. We show that readers engage in a series of pragmatic strategies to seek information. We evaluate question generation models based on GPT-2 and show that our model is able to generate reasonable questions.
  arXiv  Detail & Related papers  (2020-10-04T19:03:39Z)
• Stay Hungry, Stay Focused: Generating Informative and Specific Questions in Information-Seeking Conversations [41.74162467619795]
  We investigate the problem of generating informative questions in information-asymmetric conversations. To generate pragmatic questions, we use reinforcement learning to optimize an informativeness metric. We demonstrate that the resulting pragmatic questioner substantially improves the informativeness and specificity of questions generated over a baseline model.
  arXiv  Detail & Related papers  (2020-04-30T00:49:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.