A critical question about the nature of human learning is whether it is an all-or-none or a gradual, accumulative process. Associative and statistical theories of word learning rely critically on the later assumption: that the process of learning a word's meaning unfolds over time. That is, learning the correct referent for a word involves the accumulation of partial knowledge across multiple instances. Some theories also make an even stronger claim: Partial knowledge of one word–object mapping can speed up the acquisition of other word–object mappings. We present three experiments that test and verify these claims by exposing learners to two consecutive blocks of cross-situational learning, in which half of the words and objects in the second block were those that participants failed to learn in Block 1. In line with an accumulative account, Re-exposure to these mis-mapped items accelerated the acquisition of both previously experienced mappings and wholly new word–object mappings. But how does partial knowledge of some words speed the acquisition of others? We consider two hypotheses. First, partial knowledge of a word could reduce the amount of information required for it to reach threshold, and the supra-threshold mapping could subsequently aid in the acquisition of new mappings. Alternatively, partial knowledge of a word's meaning could be useful for disambiguating the meanings of other words even before the threshold of learning is reached. We construct and compare computational models embodying each of these hypotheses and show that the latter provides a better explanation of the empirical data.

A growing set of data show that adults are quite good at accumulating statistical evidence across individually ambiguous learning contexts with multiple novel words and multiple novel objects (Yu and Smith, 2007; Fitneva and Christiansen, 2011; Kachergis et al., 2012; Yurovsky et al., under resubmission); experimental studies also indicate that infants and young children do this kind of learning as well (Smith and Yu, 2008; Vouloumanos and Werker, 2009). The present study provides evidence for the operation of selective attention in the course of cross-situational learning with two main goals. The first was to show that selective attention is critical for the underlying mechanisms that support successful cross-situational learning. The second one was to test whether an associative mechanism with selective attention can explain momentary gaze data in cross-situational learning. Toward these goals, we collected eye movement data from participants when they engaged in a cross-situational statistical learning task. Various gaze patterns were extracted, analyzed and compared between strong learners who acquired more word-referent pairs through training, and average and weak learners who learned fewer pairs. Fine-grained behavioral patterns from gaze data reveal how learners control their attention after hearing a word, how they selectively attend to individual objects which compete for attention within a learning trial, and how statistical evidence is accumulated trial by trial, and integrated across words, across objects, and across word–object mappings. Taken together, those findings from eye movements provide new evidence on the real-time statistical learning mechanisms operating in the human cognitive system.

In this paper we propose a sequential pattern mining method to analyze multimodal data streams using a quantitative temporal approach. While the existing algorithms can only find sequential orders of temporal events, this paper presents a new temporal data mining method focusing on extracting exact timings and durations of sequential patterns extracted from multiple temporal event streams. We present our method with its application to the detection and extraction of human sequential behavioral patterns over multiple multimodal data streams in human-robot interactions. Experimental results confirmed the feasibility and quality of our proposed pattern mining algorithm, and suggested a quantitative data-driven way to ground social interactions in a manner that has never been achieved before.

Multimodal interaction in everyday life seems so effortless. However, a closer look reveals that such interaction is indeed complex and comprises multiple levels of coordination, from high-level linguistic exchanges to low-level couplings of momentary bodily movements both within an agent and across multiple interacting agents. A better understanding of how these multimodal behaviors are coordinated can provide insightful principles to guide the development of intelligent multimodal interfaces. In light of this, we propose and implement a research framework in which human participants interact with a virtual agent in a virtual environment. Our platform allows the virtual agent to keep track of the user's gaze and hand movements in real time, and adjust his own behaviors accordingly. An experiment is designed and conducted to investigate adaptive user behaviors in a human-agent joint attention task. Multimodal data streams are collected in the study including speech, eye gaze, hand and head movements from both the human user and the virtual agent, which are then analyzed to discover various behavioral patterns. Those patterns show that human participants are highly sensitive to momentary multimodal behaviors generated by the virtual agent and they rapidly adapt their behaviors accordingly. Our results suggest the importance of studying and understanding real-time adaptive behaviors in human-computer multimodal interactions.

Abstract. A better understanding of the human user's expectations and sensitivities to the real-time behavior generated by virtual agents can provide insightful empirical data and infer useful principles to guide the design of intelligent virtual agents. In light of this, we propose and implement a research framework to systematically study and evaluate different important aspects of multimodal real-time interactions between humans and virtual agents. Our platform allows the virtual agent to keep track of the user's gaze and hand movements in real time, and adjust his own behaviors accordingly. Multimodal data streams are collected in human-avatar interactions including speech, eye gaze, hand and head movements from both the human user and the virtual agent, which are then used to discover fine-grained behavioral patterns in human-agent interactions. We present a pilot study based on the proposed framework as an example of the kinds of research questions that can be rigorously addressed and answered. This first study investigating human-agent joint attention reveals promising results about the role and functioning of joint attention in human-avatar interactions.