Task-oriented Dialog (ToD) systems have to solve multiple subgoals to accomplish user goals, whereas feedback is often obtained only at the end of the dialog. In this work, we propose SUIT (SUbgoal-aware ITerative Training), an iterative training approach for improving ToD systems. We sample dialogs from the model we aim to improve and determine subgoals that contribute to dialog success using distant supervision to obtain high quality training samples. We show how this data improves supervised fine-tuning or, alternatively, preference learning results. Performance improves when applying these steps over several iterations: SUIT reaches new state-of-the-art performance on a popular ToD benchmark.

Self-rationalizing models that also generate a free-text explanation for their predicted labels are an important tool to build trustworthy AI applications. Since generating explanations for annotated labels is a laborious and costly process, recent models rely on large pretrained language models (PLMs) as their backbone and few-shot learning. In this work we explore a self-training approach leveraging both labeled and unlabeled data to further improve few-shot models, under the assumption that neither human written rationales nor annotated task labels are available at scale. We introduce a novel dual-teacher learning framework, which learns two specialized teacher models for task prediction and rationalization using self-training and distills their knowledge into a multi-tasking student model that can jointly generate the task label and rationale. Furthermore, we formulate a new loss function, Masked Label Regularization(MLR) which promotes explanations to be strongly conditioned on predicted labels. Evaluation on three public datasets demonstrate that the proposed methods are effective in modeling task labels and generating faithful rationales.

Imbalanced data distribution is a practical and common challenge in building production-level machine learning (ML) models in industry, where data usually exhibits long-tail distributions. For instance, in virtual AI Assistants, such as Google Assistant, Amazon Alexa and Apple Siri, the “play music” or “set timer” utterance is exposed to an order of magnitude more traffic than other skills. This can easily cause trained models to overfit to the majority classes, categories or intents, lead to model miscalibration. The uncalibrated models output unreliable (mostly overconfident) predictions, which are at high risk of affecting downstream decision-making systems. In this work, we study the calibration of production models in the industry use-case of predicting product return reason codes in customer service conversations of an online retail store; The returns reasons also exhibit class imbalance. To alleviate the resulting miscalibration in the production ML model, we streamline the model development and deployment using focal loss (CITATION).We empirically show the effectiveness of model training with focal loss in learning better calibrated models, as compared to standard cross-entropy loss. Better calibration, in turn, enables better control of the precision-recall trade-off for the models deployed in production.

Task-oriented dialogue systems in industry settings need to have high conversational capability, be easily adaptable to changing situations and conform to business constraints. This paper describes a 3-step procedure to develop a conversational model that satisfies these criteria and can efficiently scale to rank a large set of response candidates. First, we provide a simple algorithm to semi-automatically create a high-coverage template set from historic conversations without any annotation. Second, we propose a neural architecture that encodes the dialogue context and applicable business constraints as profile features for ranking the next turn. Third, we describe a two-stage learning strategy with self-supervised training, followed by supervised fine-tuning on limited data collected through a human-in-the-loop platform. Finally, we describe offline experiments and present results of deploying our model with human-in-the-loop to converse with live customers online.

We present the Multilingual Amazon Reviews Corpus (MARC), a large-scale collection of Amazon reviews for multilingual text classification. The corpus contains reviews in English, Japanese, German, French, Spanish, and Chinese, which were collected between 2015 and 2019. Each record in the dataset contains the review text, the review title, the star rating, an anonymized reviewer ID, an anonymized product ID, and the coarse-grained product category (e.g., ‘books’, ‘appliances’, etc.) The corpus is balanced across the 5 possible star ratings, so each rating constitutes 20% of the reviews in each language. For each language, there are 200,000, 5,000, and 5,000 reviews in the training, development, and test sets, respectively. We report baseline results for supervised text classification and zero-shot cross-lingual transfer learning by fine-tuning a multilingual BERT model on reviews data. We propose the use of mean absolute error (MAE) instead of classification accuracy for this task, since MAE accounts for the ordinal nature of the ratings.

Automatically understanding the plot of novels is important both for informing literary scholarship and applications such as summarization or recommendation. Various models have addressed this task, but their evaluation has remained largely intrinsic and qualitative. Here, we propose a principled and scalable framework leveraging expert-provided semantic tags (e.g., mystery, pirates) to evaluate plot representations in an extrinsic fashion, assessing their ability to produce locally coherent groupings of novels (micro-clusters) in model space. We present a deep recurrent autoencoder model that learns richly structured multi-view plot representations, and show that they i) yield better micro-clusters than less structured representations; and ii) are interpretable, and thus useful for further literary analysis or labeling of the emerging micro-clusters.

To create the first Hungarian WSD corpus, 39 suitable word form samples were selected for the purpose of word sense disambiguation. Among others, selection criteria required the given word form to be frequent in Hungarian language usage, and to have more than one sense considered frequent in usage. HNC and its Heti Világgazdaság subcorpus provided the basis for corpus text selection. This way, each sample has a relevant context (whole article), and information on the lemma, POS-tagging and automatic tokenization is also available. When planning the corpus, 300-500 samples of each word form were to be annotated. This size makes it possible that the subcorpora prepared for the individual word forms can be compared to data available for other languages. However, the finalized database also contains unannotated samples and samples with single annotation, which were annotated only by one of the linguists. The corpus follows the ACL’s SensEval/SemEval WSD tasks format. The first version of the corpus was developed within the scope of the project titled The construction Hungarian WordNet Ontology and its application in Information Extraction Systems (Hatvani et al., 2007). The corpus “ for research and educational purposes” is available and can be downloaded free of charge.

A highly accurate Named Entity (NE) corpus for Hungarian that is publicly available for research purposes is introduced in the paper, along with its main properties. The results of experiments that apply various Machine Learning models and classifier combination schemes are also presented to serve as a benchmark for further research based on the corpus. The data is a segment of the Szeged Corpus (Csendes et al., 2004), consisting of short business news articles collected from MTI (Hungarian News Agency, www.mti.hu). The annotation procedure was carried out paying special attention to annotation accuracy. The corpus went through a parallel annotation phase done by two annotators, resulting in a tagging with inter-annotator agreement rate of 99.89%. Controversial taggings were collected and discussed by the two annotators and a linguist with several years of experience in corpus annotation. These examples were tagged following the decision they made together, and finally all entities that had suspicious or dubious annotations were collected and checked for consistency. We consider the result of this correcting process virtually be free of errors. Our best performing Named Entity Recognizer (NER) model attained an accuracy of 92.86% F measure on the corpus.