Despite the recent progress in language generation models, their outputs may not always meet user expectations. In this work, we study whether informational feedback in natural language can be leveraged to improve generation quality and user preference alignment. To this end, we consider factual consistency in summarization, the quality that the summary should only contain information supported by the input documents, as the user-expected preference. We collect a high-quality dataset, DeFacto, containing human demonstrations and informational natural language feedback consisting of corrective instructions, edited summaries, and explanations with respect to the factual consistency of the summary. Using our dataset, we study three natural language generation tasks: (1) editing a summary by following the human feedback, (2) generating human feedback for editing the original summary, and (3) revising the initial summary to correct factual errors by generating both the human feedback and edited summary. We show that DeFacto can provide factually consistent human-edited summaries and further insights into summarization factual consistency thanks to its informational natural language feedback. We further demonstrate that fine-tuned language models can leverage our dataset to improve the summary factual consistency, while large language models lack the zero-shot learning ability in our proposed tasks that require controllable text generation.

Summarization models often generate text that is poorly calibrated to quality metrics because they are trained to maximize the likelihood of a single reference (MLE). To address this, recent work has added a calibration step, which exposes a model to its own ranked outputs to improve relevance or, in a separate line of work, contrasts positive and negative sets to improve faithfulness. While effective, much of this work has focused on how to generate and optimize these sets. Less is known about why one setup is more effective than another. In this work, we uncover the underlying characteristics of effective sets. For each training instance, we form a large, diverse pool of candidates and systematically vary the subsets used for calibration fine-tuning. Each selection strategy targets distinct aspects of the sets, such as lexical diversity or the size of the gap between positive and negatives. On three diverse scientific long-form summarization datasets (spanning biomedical, clinical, and chemical domains), we find, among others, that faithfulness calibration is optimal when the negative sets are extractive and more likely to be generated, whereas for relevance calibration, the metric margin between candidates should be maximized and surprise--the disagreement between model and metric defined candidate rankings--minimized. Code to create, select, and optimize calibration sets is available at https://github.com/griff4692/calibrating-summaries

We consider the problem of scaling automated suggested replies for Outlook email system to multiple languages. Faced with increased compute requirements and low resources for language expansion, we build a single universal model for improving the quality and reducing run-time costs of our production system. However, restricted data movement across regional centers prevents joint training across languages. To this end, we propose a multi-task continual learning framework, with auxiliary tasks and language adapters to learn universal language representation across regions. The experimental results show positive cross-lingual transfer across languages while reducing catastrophic forgetting across regions. Our online results on real user traffic show significant gains in CTR and characters saved, as well as 65% training cost reduction compared with per-language models. As a consequence, we have scaled the feature in multiple languages including low-resource markets.

We consider the problem of diversifying automated reply suggestions for a commercial instant-messaging (IM) system (Skype). Our conversation model is a standard matching based information retrieval architecture, which consists of two parallel encoders to project messages and replies into a common feature representation. During inference, we select replies from a fixed response set using nearest neighbors in the feature space. To diversify responses, we formulate the model as a generative latent variable model with Conditional Variational Auto-Encoder (M-CVAE). We propose a constrained-sampling approach to make the variational inference in M-CVAE efficient for our production system. In offline experiments, M-CVAE consistently increased diversity by ~30-40% without significant impact on relevance. This translated to a 5% gain in click-rate in our online production system.