The context of modern smart voice assistants is often multi-modal, where images, audio and video content are consumed by users simultaneously. In such a setup, co-reference resolution is especially challenging, and runs across modalities and dialogue turns. We explore the problem of multi-modal co-reference resolution in multi-turn dialogues and quantify the performance of multi-modal LLMs on a specially curated dataset of long, image-interleaved conversations between a voice assistant and human in a shopping use case. We propose a custom architecture for multi-modal embedding alignment using a novel parameter augmentation technique. Our proposed Parameter Augmented LLM approach shows a 4.9% absolute F1 improvement above a cross-attention baseline while reducing the number of parameters being trained by 4x.

Large language models incorporate world knowledge and present breakthrough performances on zero-shot learning. However, these models capture societal bias (e.g., gender or racial bias) due to bias during the training process which raises ethical concerns or can even be potentially harmful. The issue is more pronounced in multi-modal settings, such as image captioning, as images can also add onto biases (e.g., due to historical non-equal representation of genders in different occupations). In this study, we investigate the removal of potentially problematic knowledge from multi-modal models used for image captioning. We relax the gender bias issue in captioning models by degenderizing generated captions through the use of a simple linear mask, trained via adversarial training. Our proposal makes no assumption on the architecture of the model and freezes the model weights during the procedure, which also enables the mask to be turned off. We conduct experiments on COCO caption datasets using our masking solution. The results suggest that the proposed mechanism can effectively mask the targeted biased knowledge, by replacing more than 99% gender words with neutral ones, and maintain a comparable captioning quality performance with minimal (e.g., -1.4 on BLEU4 and ROUGE) impact to accuracy metrics.

In commercial dialogue systems, the Spoken Language Understanding (SLU) component tends to have numerous domains thus context is needed to help resolve ambiguities. Previous works that incorporate context for SLU have mostly focused on domains where context is limited to a few minutes. However, there are domains that have related context that could span up to hours and days. In this paper, we propose temporal representations that combine wall-clock second difference and turn order offset information to utilize both recent and distant context in a novel large-scale setup. Experiments on the Contextual Domain Classification (CDC) task with various encoder architectures show that temporal representations combining both information outperforms only one of the two. We further demonstrate that our contextual Transformer is able to reduce 13.04% of classification errors compared to a non-contextual baseline. We also conduct empirical analyses to study recent versus distant context and opportunities to lower deployment costs.

Currently, in spoken language understanding (SLU) systems, the automatic speech recognition (ASR) module produces multiple interpretations (or hypotheses) for the input audio signal and the natural language understanding (NLU) module takes the one with the highest confidence score for domain or intent classification. However, the interpretations can be noisy, and solely relying on one interpretation can cause information loss. To address the problem, many research works attempt to rerank the interpretations for a better choice while some recent works get better performance by integrating all the hypotheses during prediction. In this paper, we follow the way of integrating hypotheses but strengthen the training mode by involving more tasks, some of which may be not in existing tasks of NLU but relevant, via multi-task learning or transfer learning. Moreover, we propose the Hierarchical Attention Mechanism (HAM) to further improve the performance with the acoustic-model features like confidence scores, which are ignored in the current hypotheses integration models. The experimental results show that compared to the standard estimation with one hypothesis, the multi-task learning with HAM can improve the domain and intent classification by relatively 19% and 37%, which are much higher than improvements with current integration or reranking methods. To illustrate the cause of improvements brought by our model, we decode the hidden representations of some utterance examples and compare the generated texts with hypotheses and transcripts. The comparison shows that our model could recover the transcription by integrating the fragmented information among hypotheses and identifying the frequent error patterns of the ASR module, and even rewrite the query for a better understanding, which reveals the characteristic of multi-task learning of broadcasting knowledge.

Virtual adversarial training (VAT) is a powerful technique to improve model robustness in both supervised and semi-supervised settings. It is effective and can be easily adopted on lots of image classification and text classification tasks. However, its benefits to sequence labeling tasks such as named entity recognition (NER) have not been shown as significant, mostly, because the previous approach can not combine VAT with the conditional random field (CRF). CRF can significantly boost accuracy for sequence models by putting constraints on label transitions, which makes it an essential component in most state-of-the-art sequence labeling model architectures. In this paper, we propose SeqVAT, a method which naturally applies VAT to sequence labeling models with CRF. Empirical studies show that SeqVAT not only significantly improves the sequence labeling performance over baselines under supervised settings, but also outperforms state-of-the-art approaches under semi-supervised settings.

Adversarial training (AT) has shown strong regularization effects on deep learning algorithms by introducing small input perturbations to improve model robustness. In language tasks, adversarial training brings word-level robustness by adding input noise, which is beneficial for text classification. However, it lacks sufficient contextual information enhancement and thus is less useful for sequence labelling tasks such as chunking and named entity recognition (NER). To address this limitation, we propose masked adversarial training (MAT) to improve robustness from contextual information in sequence labelling. MAT masks or replaces some words in the sentence when computing adversarial loss from perturbed inputs and consequently enhances model robustness using more context-level information. In our experiments, our method shows significant improvements on accuracy and robustness of sequence labelling. By further incorporating with ELMo embeddings, our model achieves better or comparable results to state-of-the-art on CoNLL 2000 and 2003 benchmarks using much less parameters.