arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.03284cs.CLcs.AIcs.LG
收藏

成交,还是不成交(或者谁知道呢)?利用大型语言模型预测对话中的不确定性

Deal, or no deal (or who knows)? Forecasting Uncertainty in Conversations using Large Language Models

Northeastern University(东北大学);Allen Institute for AI(艾伦人工智能研究所);Samaya AI(萨马亚人工智能公司)

Anthony Sicilia, Hyunwoo Kim, Khyathi Raghavi Chandu, Malihe Alikhani, Jack Hessel

AI总结:

本文提出FortUne Dial任务,扩展对话预测以纳入不确定性感知评估,允许弃权(不执行),并通过微调策略校准语言模型的不确定性表示,使小模型在八个谈判语料库上能与大10倍的模型竞争。

摘要:

有效的对话者会考虑他人不确定的目标、信念和情感。但即使是最优秀的人类对话者也无法完美预测对话的轨迹。语言模型能在多大程度上代表对话中固有的不确定性?我们提出了FortUne Dial,这是对长期存在的“对话预测”任务的扩展:除了准确性之外,评估还采用不确定性感知指标,从而有效地允许在个别实例上弃权(不执行)。我们研究了语言模型可能代表结果不确定性的两种方式(内部使用分数和直接使用标记),并提出了微调策略以改进这两种表示的校准。在八个困难的谈判语料库上的实验表明,我们提出的微调策略(一种传统监督策略和一种离策略强化学习策略)可以将较小的开源模型校准到与比其大10倍的预训练模型竞争的水平。

英文摘要:

Effective interlocutors account for the uncertain goals, beliefs, and emotions of others. But even the best human conversationalist cannot perfectly anticipate the trajectory of a dialogue. How well can language models represent inherent uncertainty in conversations? We propose FortUne Dial, an expansion of the long-standing "conversation forecasting" task: instead of just accuracy, evaluation is conducted with uncertainty-aware metrics, effectively enabling abstention on individual instances. We study two ways in which language models potentially represent outcome uncertainty (internally, using scores and directly, using tokens) and propose fine-tuning strategies to improve calibration of both representations. Experiments on eight difficult negotiation corpora demonstrate that our proposed fine-tuning strategies (a traditional supervision strategy and an off-policy reinforcement learning strategy) can calibrate smaller open-source models to compete with pre-trained models 10x their size.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑