Unsupervised Domain Adaptation for Dialogue Sequence Labeling Based on Hierarchical Adversarial Training

Shota Orihashi, Mana Ihori, Tomohiro Tanaka, Ryo Masumura
<span title="2020-10-25">2020</span> <i title="ISCA"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/trpytsxgozamtbp7emuvz2ypra" style="color: black;">Interspeech 2020</a> </i> &nbsp;
This paper presents a novel unsupervised domain adaptation method for dialogue sequence labeling. Dialogue sequence labeling is a supervised learning task that estimates labels for each utterance in the given dialogue document, and is useful for many applications such as topic segmentation and dialogue act estimation. Accurate labeling often requires a large amount of labeled training data, but it is difficult to collect such data every time we need to support a new domain, such as contact
more &raquo; ... rs in a new business field. In order to solve this difficulty, we propose an unsupervised domain adaptation method for dialogue sequence labeling. Our key idea is to construct dialogue sequence labeling using labeled source domain data and unlabeled target domain data so as to remove domain dependencies at utterance-level and dialogue-level contexts. The proposed method adopts hierarchical adversarial training; two domain adversarial networks, an utterance-level context independent network and a dialogue-level context dependent network, are introduced for improving domain invariance in the dialogue sequence labeling. Experiments on Japanese simulated contact center dialogue datasets demonstrate the effectiveness of the proposed method. Index Terms: dialogue sequence labeling, unsupervised domain adaptation, hierarchical adversarial training 2. Related work 2.1. Utterance-level dialogue sequence labeling Utterance-level sequence labeling is used for topic segmentation and dialogue act estimation [7] [8] [9] [10] [11] [12] [13] [14] . Hierarchical recurrent neural networks based on token units and utterance units are often used to efficiently capture short-term contexts between tokens and long-term contexts between utterances. In this paper, we focus on the utterance-level sequence labeling of hierarchical recurrent neural networks specialized for conversation documents [12] . In order to eliminate the need to collect a large
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.21437/interspeech.2020-2010">doi:10.21437/interspeech.2020-2010</a> <a target="_blank" rel="external noopener" href="https://dblp.org/rec/conf/interspeech/OrihashiITM20.html">dblp:conf/interspeech/OrihashiITM20</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/xlkbbfo7jnf77eripy42h65hgu">fatcat:xlkbbfo7jnf77eripy42h65hgu</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20201211023541/https://www.isca-speech.org/archive/Interspeech_2020/pdfs/2010.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/47/66/4766ca24a0c8cdbf3424526b09f1f668639a1b19.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.21437/interspeech.2020-2010"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> Publisher / doi.org </button> </a>