Speaker-change Aware CRF for Dialogue Act Classification
This is an incremental improvement for dialogue systems, addressing a specific bottleneck in sequence labeling.
The paper tackles the problem of Dialogue Act classification by modifying the CRF layer to incorporate speaker-change information, which was previously ignored, and shows improved performance on the SwDA corpus with wide margins for some labels.
Recent work in Dialogue Act (DA) classification approaches the task as a sequence labeling problem, using neural network models coupled with a Conditional Random Field (CRF) as the last layer. CRF models the conditional probability of the target DA label sequence given the input utterance sequence. However, the task involves another important input sequence, that of speakers, which is ignored by previous work. To address this limitation, this paper proposes a simple modification of the CRF layer that takes speaker-change into account. Experiments on the SwDA corpus show that our modified CRF layer outperforms the original one, with very wide margins for some DA labels. Further, visualizations demonstrate that our CRF layer can learn meaningful, sophisticated transition patterns between DA label pairs conditioned on speaker-change in an end-to-end way. Code is publicly available.