Incorporating Instructional Prompts into a Unified Generative Framework for Joint Multiple Intent Detection and Slot Filling

Proceedings of the 29th International Conference on Computational Linguistics, pages 7203–7208 October 12–17, 2022. 7203 Incorporating Instructional Prompts into A Unified Generative Framework for Joint Multiple Intent Detection and Slot Filling Yangjun Wu, Han Wang, Dongxiang Zhang∗, Gang Chen, Hao Zhang Zhejiang University {yjwu, zh}@zjuici.com, {22021066, zhangdongxiang,cg }@zju.edu.cn Abstract The joint multiple Intent Detection (ID) and Slot Filling (SF) is a significant challenge in spoken language understanding. Because the slots in an utterance may relate to multi-intents, most existing approaches focus on utilizing task-specific components to capture the re- lations between intents and slots. The cus- tomized networks restrict models from mod- eling commonalities between tasks and gen- eralization for broader applications. To ad- dress the above issue, we propose a Unified Generative framework (UGEN) based on a prompt-based paradigm, and formulate the task as a question-answering problem. Specifically, we design 5-type templates as instructional prompts, and each template includes a ques- tion that acts as the driver to teach UGEN to grasp the paradigm, options that list the candidate intents or slots to reduce the an- swer search space, and the context denotes original utterance. Through the instructional prompts, UGEN is guided to understand in- tents, slots, and their implicit correlations. On two popular multi-intent benchmark datasets, experimental results demonstrate that UGEN achieves new SOTA performances on full-data and surpasses the baselines by a large margin on 5-shot (28.1%) and 10-shot (23%) scenarios, which verify that UGEN is robust and effective. Our code will be publicly available at https: //github.com/Young1993/UGEN 1 Introduction In task-oriented dialogue systems, spoken language understanding (SLU) is a crucial component that aims to understand users’ queries and use a seman- tic frame to represent users’ requirements. The semantic frame usually contains intents and slot names (Tur and De Mori, 2011). Recently, multi- ple intent SLU has attracted lots of attention (Liu and Lane, 2016; E et al., 2019; Weld et al., 2021; ∗* Corresponding author Figure 1: The semantic frame. An example from MixS- NIPS dataset(Coucke et al., 2018; Qin et al., 2020). Gangadharaiah and Narayanaswamy, 2019) due to the wide variety of practical application scenarios. Considering the example shown in Figure 1, the models are expected to identify the intents (AddTo- PlayList and RateBook) and the slot values with tags for the utterance. Current works (Qin et al., 2019, 2020; Ding et al., 2021; Qin et al., 2021; Chen et al., 2021a) usually treat Intent Detection (ID) as a classification task and Slot Filling (SF) as a sequence labeling task. The task-specific com- ponents are employed by current works to capture the connection or interaction between ID and SF, which achieve fine-grained multi-intent informa- tion integration for slot filling and obtain remark- able success. In this paper, we’re interested in exploiting a united paradigm to handle the task instead of cus- tomized networks. Prompt-learning (Liu et al., 2021; Jin et al., 2022) is a novel paradigm, which replaces the "pre-train, fine-tune" procedure with "pre-train, prompt, and predict" analogous to origi- nal pre-training language models …

Content truncated to 500 words