A Student-Teacher Architecture for Dialog Domain Adaptation Under the Meta-Learning Setting

Abstract

Numerous new dialog domains are being created every day while collecting data for these domains is extremely costly since it involves human interactions. Therefore, it is essential to develop algorithms that can adapt to different domains efficiently when building data-driven dialog models. Most recent research on domain adaption focuses on giving the model a better initialization, rather than optimizing the adaptation process. We propose an efficient domain adaptive task-oriented dialog system model, which incorporates a meta-teacher model to emphasize the different impacts between generated tokens with respect to the context. We first train our base dialog model and meta-teacher model adversarially in a meta-learning setting on rich-resource domains. The meta-teacher learns to quantify the importance of tokens under different contexts across different domains. During adaptation, the meta-teacher guides the dialog model to focus on important tokens in order to achieve better adaptation efficiency. We evaluate our model on two multi-domain datasets, MultiWOZ and Google Schema-Guided Dialogue, and achieve state-of-the-art performance.

HOP to the Next Tasks and Domains for Continual Learning in NLP

“…Early works tackled lifelong learning (with no knowledge preservation, hence no CF) for sentiment analysis (Carlson et al 2010;Silver, Yang, and Li 2013;Ruvolo and Eaton 2013;Chen, Ma, and Liu 2015;Wang et al 2019;Qin, Hu, and Liu 2020;Wang et al 2018). Recent works have dealt with CF in many applications: sentiment analysis (Lv et al 2019;Ke et al 2021b;Ke, Xu, and Liu 2021), dialogue systems (Shen, Zeng, and Jin 2019;Madotto et al 2020;Qian, Wei, and Yu 2021;Chien and Chen 2021), language modeling (Sun, Ho, and Lee 2019;Chuang, Su, and Chen 2020) and learning (Li et al 2019), cross-lingual modeling (Liu et al 2020), sentence embedding (Liu, Ungar, and Sedoc 2019), machine translation (Khayrallah et al 2018;Zhan et al 2021), question answering (Greco et al 2019), named entity recognition (Monaikul et al 2021).…”

Learning a Better Initialization for Soft Prompts via Meta-Learning

“…Model-Agnostic Meta-Learning (MAML) proposed by Finn et al (2017) is a popular optimization-based meta-learning algorithm, which is adopted in various NLP tasks (e.g. Qian et al, 2021, Gu et al, 2018, Yin, 2020, Qian and Yu, 2019, Dou et al, 2019. Following MAML, works like REPTILE (Nichol et al, 2018), MetaOPT and TAML (Jamal et al, 2019) etc.…”