An important difficulty in developing spoken dialogue systems for robots is the open-ended nature of most interactions. Robotic agents must typically operate in complex, continuously changing environments which are difficult to model and do not provide any clear, predefined goal. Directly capturing this complexity in a single, large dialogue policy is thus inadequate. This paper presents a new approach which tackles the complexity of open-ended interactions by breaking it into a set of small, independent policies, which can be activated and deactivated at runtime by a dedicated mechanism. The approach is currently being implemented in a spoken dialogue system for autonomous robots.
Open-ended student answers often need to be assessed in context. However, there are not many previous works that consider context when automatically assessing student answers. Furthermore, student responses vary significantly in their explicit content and writing style which leads to a wide range of assessment scores for the same qualitative assessment category, e.g. correct answers vs. incorrect answers. In this paper, we propose an approach to assessing student answers that takes context into account and which handles variability using probabilistic Gaussian Mixture Models (GMMs). We developed the model using a recently released corpus called DT-Grade which was manually annotated, taking context into account, with four different levels of answer correctness. Our best GMM model outperforms the baseline model with a margin of 9% in terms of accuracy.
Dialogue policy optimization often obtains feedback until task completion in task-oriented dialogue systems. This is insufficient for training intermediate dialogue turns since supervision signals (or rewards) are only provided at the end of dialogues. To address this issue, reward learning has been introduced to learn from state-action pairs of an optimal policy to provide turn-by-turn rewards. This approach requires complete state-action annotations of human-to-human dialogues (i.e., expert demonstrations), which is labor intensive. To overcome this limitation, we propose a novel reward learning approach for semi-supervised policy learning. The proposed approach learns a dynamics model as the reward function which models dialogue progress (i.e., state-action sequences) based on expert demonstrations, either with or without annotations. The dynamics model computes rewards by predicting whether the dialogue progress is consistent with expert demonstrations. We further propose to learn action embeddings for a better generalization of the reward function. The proposed approach outperforms competitive policy learning baselines on MultiWOZ, a benchmark multi-domain dataset.
Training chatbots using the reinforcement learning paradigm is challenging due to high-dimensional states, infinite action spaces and the difficulty in specifying the reward function. We address such problems using clustered actions instead of infinite actions, and a simple but promising reward function based on human-likeness scores derived from human-human dialogue data. We train Deep Reinforcement Learning (DRL) agents using chitchat data in raw text---without any manual annotations. Experimental results using different splits of training data report the following. First, that our agents learn reasonable policies in the environments they get familiarised with, but their performance drops substantially when they are exposed to a test set of unseen dialogues. Second, that the choice of sentence embedding size between 100 and 300 dimensions is not significantly different on test data. Third, that our proposed human-likeness rewards are reasonable for training chatbots as long as they use lengthy dialogue histories of >=10 sentences.
Developing conversational agents to engage in complex dialogues is challenging partly because the dialogue policy needs to explore a large state-action space. In this paper, we propose a divide-and-conquer approach that discovers and exploits the hidden structure of the task to enable efficient policy learning. First, given a set of successful dialogue sessions, we present a Subgoal Discovery Network (SDN) to divide a complex goal-oriented task into a set of simpler subgoals in an unsupervised fashion. We then use these subgoals to learn a hierarchical policy which consists of 1) a top-level policy that selects among subgoals, and 2) a low-level policy that selects primitive actions to accomplish the subgoal. We exemplify our method by building a dialogue agent for the composite task of travel planning. Experiments with simulated and real users show that an agent trained with automatically discovered subgoals performs competitively against an agent with human-defined subgoals, and significantly outperforms an agent without subgoals. Moreover, we show that learned subgoals are human comprehensible.