Dominique Knutsen

Capturing egocentric biases in reference reuse during collaborative dialogue

Words that are produced aloud—and especially self-produced ones—are remembered better than words that are not, a phenomenon labeled the production effect in the field of memory research. Two experiments were conducted to determine whether this effect can be generalized to dialogue, and how it might affect dialogue management. Triads (Exp. 1) or dyads (Exp. 2) of participants interacted to perform a collaborative task. Analyzing reference reuse during the interaction revealed that the participants were more likely to reuse the references that they had presented themselves, on the one hand, and those that had been accepted through verbatim repetition, on the other. Analyzing reference recall suggested that the greater accessibility of self-presented references was only transient. Moreover, among partner-presented references, those discussed while the participant had actively taken part in the conversation were more likely to be recalled than those discussed while the participant had been inactive. These results contribute to a better understanding of how individual memory processes might contribute to collaborative dialogue.

Conceptual match as a determinant of reference reuse in dialogue.

As speakers interact, they add references to their common ground, which they can then reuse to facilitate listener comprehension. However, all references are not equally likely to be reused. The purpose of this study was to shed light on how the speakers’ conceptualizations of the referents under discussion affect reuse (along with a generation effect in memory documented in previous studies on dialogic reuse). Two experiments were conducted in which participants interactively added references to their common ground. From each participant’s point of view, these references either did or did not match their own conceptualization of the referents discussed, and were either self- or partner-generated. Although self-generated references were more readily accessible in memory than partner-generated ones (Experiment 1), reference reuse was mainly guided by conceptualization (Experiment 2). These results are in line with the idea that several different cues (conceptual match, memory accessibility) constrain reference reuse in dialogue.

Explicit feedback from users attenuates memory biases in human-system dialogue

In human–human dialogue, the way in which a piece of information is added to the partners’ common ground (i.e., presented and accepted) constitutes an important determinant of subsequent dialogue memory. The aim of this study was to determine whether this is also the case in human-system dialogue. An experiment was conducted in which naive participants and a simulated dialogue system took turns to present references to various landmarks featured on a list. The kind of feedback used to accept these references (verbatim repetition vs. implicit acceptance) was manipulated. The participants then performed a recognition test during which they attempted to identify the references mentioned previously. Self-presented references were recognised better than references presented by the system; however, such presentation bias was attenuated when the initial presentation of these references was followed by verbatim repetition. Implications for the design of automated dialogue systems are discussed.

Generating References in Naturalistic Face-to-Face and Phone-Mediated Dialog Settings

During dialog, references are presented, accepted, and potentially reused (depending on their accessibility in memory). Two experiments were conducted to examine reuse in a naturalistic setting (a walk in a familiar environment). In Experiment 1, where the participants interacted face to face, self-presented references and references accepted through verbatim repetition were reused more. Such biases persisted after the end of the interaction. In Experiment 2, where the participants interacted over the phone, reference reuse mainly depended on whether the participant could see the landmarks being referred to, although this bias seemed to be only transient. Consistent with the memory-based approach to dialog, these results shed light on how differences in accessibility in memory (due to how these references were initially added to the common ground or the media used) affect the unfolding of the interaction.

Spoilt for choice: initially considering several referential expressions affects subsequent referential decisions

ABSTRACT In dialogue, speakers jointly decide how to refer to the referents under discussion. In some cases, several different referential expressions are considered before the partners can decide which one they prefer; this work examined how doing so affects subsequent referential expression reuse. Pairs of participants came up with suitable referential expressions for Tangram figures they were shown. They then referred to the same figures again during a matching task which was performed either with the same partner or a different partner. The main finding was that the preferred referential expression was less likely to be reused when several referential expressions were initially considered. This effect could not be attributed to a generation effect or to some referential expressions being a better match for the Tangram figures than others. These findings offer a better understanding of how the initial contribution of a reference shapes subsequent referential decisions through ordinary memory functioning.

Reducing user linguistic variability in speech interaction through lexical and syntactic priming

Motivation -- To reduce user linguistic variability in human-system interaction. Research approach -- An experiment was conducted in which 72 participants interacted over the phone with a simulated natural language dialogue system. The main manipulation concerned the lexical content and the structure of the message prompts. Findings/Design -- The results confirm that users align with the system on the lexical and structural levels in human-system dialogue. However, the strength of the syntactic alignment depends on the content of the prime. Research limitations/Implications -- This experiment should be replicated user a greater number of different prime system messages. Originality/Value -- By manipulating prime content, this study allows investigating alignment strength as a function of word frequency and user preferences. Take-away message -- Lexical and syntactic priming can be used to reduce user linguistic variability in human-system interaction, but the strength of these phenomena depends on the content of the prime.

The influence of conceptual (mis)match on collaborative referring in dialogue

When two dialogue partners need to refer to something, they jointly negotiate which referring expression should be used. If needed, the chosen referring expression is then reused throughout the interaction, which potentially has a direct, positive impact on subsequent communication. The purpose of this study was to determine if the way in which the partners view, or conceptualise, the referent under discussion, affects referring expression negotiation and subsequent communication. A matching task was preceded by an individual task during which participants were required to describe their conceptualisations of abstract tangram pictures. The results revealed that participants found it more difficult to converge on single referring expression during the matching task when they initially held different conceptualisations of the pictures. This had a negative impact on the remainder of the task. These findings are discussed in light of the shared versus mutual knowledge distinction, highlighting how the former directly contributes to the formation of the latter.

I remember emotional content better, but I’m struggling to remember who said it!

The joint impact of emotion and production on conversational memory was examined in two experiments where pairs of participants took turns producing verbal information. They were instructed to produce out loud sentences based on either neutral or emotional (Experiment 1: negative; Experiment 2: positive) words. Each participant was then asked to recall as many words as possible (content memory) and to indicate who had produced each word (reality monitoring). The analyses showed that both self-production and emotion boost content memory, although emotion also impairs reality monitoring. This study sheds light on how both factors (emotion and production) may constrain language interaction memory through information saliency.

How does the representation of the computer evolve during human-computer dialogue?

Motivation -- To better understand how the representation of the computers capabilities and needs built by the user evolves through interaction. Research approach -- An experiment is being conducted where participants interact with a simulated dialogue system. We look at the order in which users produce information as an indicator of how the computers needs and capabilities are perceived. Findings/Design -- So far, the results show that the representation of the computers capabilities evolves. Users perceive the system more like a human as the interaction unfolds. Thus, if the aim is to design a human-like system which elicits collaborative behaviour from the user, starting by presenting the system as collaborative could benefit the interaction. Research limitations/Implications -- In this study, interactions were limited to reference production. Originality/Value -- This experiment looks at information production as a way of determining how the representation of the computers needs evolves as the user has the opportunity to interact with the system. Take away message -- The way in which a user perceives a dialogue system evolves as the interaction unfolds. This affects knowledge mobilisation by the user: the more he or she interacts with the system, the more the latter is perceived as a collaborative partner. This can be exploited for system design.