Representation Learning for Conversational Data using Discourse Mutual Information Maximization

Santra, Bishal; Roychowdhury, Sumegh; Mandal, Aishik; Gurram, Vasu; Naik, Ajit Kumar; Gupta, Manish; Goyal, Pawan

doi:10.48550/arxiv.2112.05787

Cited by 1 publication

(3 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Such a measure is learned by retrieval models like Chen and Wang [2019], Henderson et al [2020]. Santra et al [2021] The main idea of CORAL loss is to optimize a measure of the compatibility between the context and a candidate response. For our implementation, we have used a response retrieval model for measuring this compatibility.…”

Section: The Coral Loss Functionmentioning

confidence: 99%

“…• Retrieval model There is also the choice of the retrieval model that is used for measuring R 3 . Here we have chosen ESIM [Chen and Wang, 2019], BERT [Devlin et al, 2019] and DMI [Santra et al, 2021] as the base models for training the response retrieval (i.e., R 3 reward) models.…”

Section: Hyperparameters Of Coralmentioning

confidence: 99%

“…Sinha et al [2020], Yeh et al [2021] have shown that the output score of retrieval models correlates strongly with human perception of dialog response quality (given that it was trained on a dataset of domain similar to the target domain of application). Further, Santra et al [2021] showed that representations learned by a response retrieval (using binary cross-entropy or a contrastive loss) capture important dialog understanding features even better than large-scale dialog generation models trained using CE.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

CORAL: Contextual Response Retrievability Loss Function for Training Dialog Generation Models

Santra¹,

Ghadia²,

Dwivedi³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Natural Language Generation (NLG) represents a large collection of tasks in the field of NLP. While many of these tasks have been tackled well by the cross-entropy (CE) loss, the task of dialog generation poses a few unique challenges for this loss function. First, CE loss assumes that for any given input, the only possible output is the one available as the ground truth in the training dataset. In general, this is not true for any task, as there can be multiple semantically equivalent sentences, each with a different surface form. This problem gets exaggerated further for the dialog generation task, as there can be multiple valid responses (for a given context) that not only have different surface forms but are also not semantically equivalent. Second, CE loss does not take the context into consideration while processing the response and, hence, it treats all ground truths with equal importance irrespective of the context. But, we may want our final agent to avoid certain classes of responses (e.g. bland, non-informative or biased responses) and give relatively higher weightage for more context-specific responses. To circumvent these shortcomings of the CE loss, in this paper, we propose a novel loss function, CORAL, that directly optimizes recently proposed estimates of human preference for generated responses. Using CORAL, we can train dialog generation models without assuming non-existence of response other than the ground-truth. Also, the CORAL loss is computed based on both the context and the response. Our experiments show that, against various large and small scale baselines, our model is able to obtain significantly higher scores for the human preference estimate. While large-scale dialog models have shown much promise and aptitude for dialog generation, it is important to continue the search for more suitable loss functions for training dialog systems. Extensive comparisons on two benchmark datasets show that the proposed methods outperform strong state-of-the-art baseline models of different sizes. We will make the code and trained model checkpoints publicly available upon publication of this paper.Preprint. Under review.

show abstract

Section: The Coral Loss Functionmentioning

confidence: 99%