Richard Shin, Christopher Lin, Sam Thomson, Charles Chen, Subhro Roy, Emmanouil Antonios Platanios, Adam Pauls, Dan Klein, Jason Eisner, Benjamin Van Durme. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.

We describe an approach to task-oriented dialogue in which dialogue state is represented as a dataflow graph. A dialogue agent maps each user utterance to a program that extends this graph. Programs include metacomputation operators for reference and revision that reuse dataflow fragments from previous turns. Our graph-based state enables the expression and manipulation of complex user intents, and explicit metacomputation makes these intents easier for learned models to predict. We introduce a new dataset, SMCalFlow, featuring complex dialogues about events, weather, places, and people. Experiments show that dataflow graphs and metacomputation substantially improve representability and predictability in these natural dialogues. Additional experiments on the MultiWOZ dataset show that our dataflow representation enables an otherwise off-the-shelf sequence-to-sequence model to match the best existing task-specific state tracking model. The SMCalFlow dataset, code for replicating experiments, and a public leaderboard are available at https://www.microsoft.com/en-us/research/project/dataflow-based-dialogue-semantic-machines .

Email triage involves going throughunhandled emails and deciding what to do with them. This familiar process can become increasingly challenging as the number of unhandled email grows. During a triage session, users commonlydefer handling emails that they cannot immediately deal with to later. These deferred emails, are often related to tasks that are postponed until the user has more time or the right information to deal with them. In this paper, through qualitative interviews and a large-scale log analysis, we study when and whatenterprise email users tend to defer. We found that users are more likely to defer emails when handling them involves replying, reading carefully, or clicking on links and attachments. We also learned that the decision to defer emails depends on many factors such as user's workload and the importance of the sender. Our qualitative results suggested that deferring is very common, and our quantitative log analysis confirms that 12% of triage sessions and 16% of daily active users had at least one deferred email on weekdays. We also discuss severaldeferral strategies such as marking emails as unread and flagging that are reported by our interviewees, and illustrate how such patterns can be also observed in user logs. Inspired by the characteristics of deferred emails and contextual factors involved in deciding if an email should be deferred, we train a classifier for predicting whether a recently triaged email is actually deferred. Our experimental results suggests that deferral can be classified with modest effectiveness. Overall, our work provides novel insights about how users handle their emails and how deferral can be modeled.

Email remains a critical channel for communicating information in both personal and work accounts. The number of emails people receive every day can be overwhelming, which in turn creates challenges for efficient information management and consumption. Having a good estimate of the significance of emails forms the foundation for many downstream tasks (e.g. email prioritization); but determining significance at scale is expensive and challenging.

Machine learning in real-world high-skew domains is difficult, because traditional strategies for crowdsourcing labeled training examples are ineffective at locating the scarce minority-class examples. For example, both random sampling and traditional active learning (which reduces to random sampling when just starting) will most likely recover very few minority-class examples. To bootstrap the machine learning process, researchers have proposed tasking the crowd with finding or generating minority-class examples, but such strategies have their weaknesses as well. They are unnecessarily expensive in well-balanced domains, and they often yield samples from a biased distribution that is unrepresentative of the one being learned.This paper extends the traditional active learning framework by investigating the problem of intelligently switching between various crowdsourcing strategies for obtaining labeled training examples in order to optimally train a classifier. We start by analyzing several such strategies (e.g., annotate an example, generate a minority-class example, etc.), and then develop a novel, skew-robust algorithm, called MB-CB, for the control problem. Experiments show that our method outperforms state-of-the-art GL-Hybrid by up to 14.3 points in F1 AUC, across various domains and class-frequency settings.

Active learning seeks to train the best classifier at the lowest annotation cost by intelligently picking the best examples to label. Traditional algorithms assume there is a single annotator and disregard the possibility of requesting additional independent annotations for a previously labeled example. However, relabeling examples is important, because all annotators make mistakes — especially crowdsourced workers, who have become a common source of training data. This paper seeks to understand the difference in marginal value between decreasing the noise of the training set via relabeling and increasing the size and diversity of the (noisier) training set by labeling new examples. We use the term re-active learning to denote this generalization of active learning. We show how traditional active learning methods perform poorly at re-active learning, present new algorithms designed for this important problem, formally characterize their behavior, and empirically show that our methods effectively make this tradeoff.

Angli Liu, Stephen Soderland, Jonathan Bragg, Christopher H. Lin, Xiao Ling, Daniel S. Weld. Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2016.

We exploit the absence of signals as informative observations in the context of providing task recommendations in crowdsourcing. Workers on crowdsourcing platforms do not provide explicit ratings about tasks. We present methods that enable a system to leverage implicit signals about task preferences. These signals include types of tasks that have been available and have been displayed, and the number of tasks workers select and complete. In contrast to previous work, we present a general model that can represent both positive and negative implicit signals. We introduce algorithms that can learn these models without exceeding the computational complexity of existing approaches. Finally, using data from a high-throughput crowdsourcing platform, we show that reasoning about both positive and negative implicit feedback can improve the quality of task recommendations.

One of the most popular uses of crowdsourcing is to provide training data for supervised machine learning algorithms. Since human annotators often make errors, requesters commonly ask multiple workers to label each example. But is this strategy always the most cost effective use of crowdsourced workers? We argue "No" --- often classifiers can achieve higher accuracies when trained with noisy "unilabeled" data. However, in some cases relabeling is extremely important. We discuss three factors that may make relabeling an effective strategy: classifier expressiveness, worker accuracy, and budget.

To ensure quality results from unreliable crowdsourced workers, task designers often construct complex workflows and aggregate worker responses from redundant runs. Frequently, they experiment with several alternative workflows to accomplish the task, and eventually deploy the one that achieves the best performance during early trials. Surprisingly, this seemingly natural design paradigm does not achieve the full potential of crowdsourcing. In particular, using a single workflow (even the best) to accomplish a task is suboptimal. We show that alternative workflows can compose synergistically to yield much higher quality output. We formalize the insight with a novel probabilistic graphical model. Based on this model, we design and implement AGENTHUNT, a POMDP-based controller that dynamically switches between these workflows to achieve higher returns on investment. Additionally, we design offline and online methods for learning model parameters. Live experiments on Amazon Mechanical Turk demonstrate the superiority of AGENTHUNT for the task of generating NLP training data, yielding up to 50% error reduction and greater net utility compared to previous methods.