A New Theoretical Framework for Curiosity for Learning in Social Contexts

TL;DR

This paper introduces a novel socio-cognitive framework for understanding how social factors influence curiosity in group learning, validated through empirical data and modeling.

Contribution

It presents the first integrated theoretical framework linking social dynamics and curiosity, supported by empirical observations and advanced modeling techniques.

Findings

Interpersonal functions have a stronger influence on curiosity than individual functions.

Validated the framework using longitudinal latent variable modeling.

Identified key multimodal behaviors associated with curiosity in social learning.

Abstract

Curiosity is a vital metacognitive skill in educational contexts. Yet, little is known about how social factors influence curiosity in group work. We argue that curiosity is evoked not only through individual, but also interpersonal activities, and present what we believe to be the first theoretical framework that articulates an integrated socio-cognitive account of curiosity based on literature spanning psychology, learning sciences and group dynamics, along with empirical observation of small-group science activity in an informal learning environment. We make a bipartite distinction between individual and interpersonal functions that contribute to curiosity, and multimodal behaviors that fulfill these functions. We validate the proposed framework by leveraging a longitudinal latent variable modeling approach. Findings confirm positive predictive relationship of the latent variables of individual and interpersonal functions on curiosity, with the interpersonal functions exercising a comparatively stronger influence. Prominent behavioral realizations of these functions are also discovered in a data-driven way. This framework is a step towards designing learning technologies that can recognize and evoke curiosity during learning in social contexts.

Figures

Tables

Table 1. Table 1: Corpus examples of behavior sequences. P1 is the child with high curiosity

Behavior Cluster	Empirical Observation (Example 1)	Empirical Observation (Example 2)
Cluster 1,2	\pbox6.5cmP1: Hey let’s..wait I have an idea [idea verbalization]
P1: Let’s see what this is, but let me just, let me just.. [proposes joint action, co-occurs with physical demonstration, initiates joint inquiry]
P2: I have no idea how to do this, but it’s making my brain think [positive attitude towards task]	\pbox6.5cmP1: So the chain has to be like this [idea verbalization with iconic gesture]
P1: How would that be? [question asking followed by orienting towards stimulus]
P1: Well, I don’t want it to break, so I want it to be about…no, let’s say half an…half an inch [causal reasoning to justify actions being taken]
Cluster 1,3	\pbox6.5cmP1: Wait we need to raise it a bit higher [making suggestions]
P1: Maybe if we put it on..Umm..this thing maybe..this is high enough? [co-occurs with joint stimulus manipulation]
P2: Why? W-Why do we need to make it that high? [disagreement and asking for evidence]	\pbox6.5cmP2: And the funnel can drop it into one of um..those things
P1: If the funnel can drop it…
P1: Okay but then..even if it hits this, then we need what is this going to hit? [challenge]
P1: Here- let- just- make sure that it’s going to hit it [followed by physical demonstration/verification]
Cluster 2,3,4	\pbox6.5cmP1: Roll off into here and go in there [hypothesis generation]
P1: Okay, so how are we going to do that? [question asking]
P2: It looks like something should hit the ball [making suggestion]	\pbox6.5cmP2: We could use this if we wanted [making suggestion]
P1: Let’s figure this quickly…so we at least have this part done [preceded by expression of surprise and followed by trying to connect multiple objects to create a more complex object]

Table 2. Table 2: A summary of coding methods used for the annotation. Detailed coding scheme for verbal behaviors can be found at http://tinyurl.com/codingschemecuriosity

Construct	Definition used to code/infer the construct	Coding method
\pbox2.5cmGround Truth
Curiosity	A strong desire to learn or know more about something or someone.	Four MTurk raters annotated each 10-sec thin slice; average ICC=0.46; used inverse-based bias correction to pick the final rating.
Verbal Behavior
1. Uncertainty	Lack of certainty about ones choices or beliefs, and is verbally expressed by language that creates an impression that something important has been said, but what is communicated is vague, misleading, evasive or ambiguous. e.g - “well maybe we should use rubberbands on the foam pieces”	\pbox3cmUsed a semi-automated annotation approach: after automatic labeling of these verbal behaviors, two trained raters (Krippendroff’s alpha $>$0.6) independently corrected machine annotated labels; average percentage of machine annotation that remained the same after human correction was 85.9 (SD=12.71).
2. Argument	A coherent series of reasons, statements, or facts intended to support or establish a point of view. e.g -“no we got to first find out the chain reactions that it can do”
3. Justification	The action of showing something to be right or reasonable by making it clear. e.g -‘wait with the momentum of going downhill it will go straight into the trap”
4. Suggestion	An idea or plan put forward for consideration. e.g - “you are adding more weight there which would make it fall down”
5. Agreement	Harmony or accordance in opinion or feeling; a position or result of agreeing. e.g - “And we put the ball in here..I hope it still works, and it goes..so it starts like that, and then we hit it” [Quote] — “Ok that works” [Response]
6. Question Asking (On-Task/Social)	Asking any kind of questions related to the task or non-task relevant aspects of the social interaction. e.g - “why do we need to make it that high?”, “do you two go to the same school?”	\pbox3cmUsed manual annotation procedure due to unavailability of existing training corpus (Krippendroff’s alpha $>$0.76 between two raters).
7. Idea Verbalization	Explicitly saying out an idea, which can be just triggered by an individual’s own actions or something that builds off of other peer’s actions. e.g - “yeah that ball isn’t heavy enough”
8. Sharing Findings	An explicit verbalization of communicating results, findings and discoveries to group members during any stage of a scientific inquiry process. e.g - “look how I’m gonna see I’m gonna trap it”
9. Hypothesis Generation	Expressing one or more different possibilities or theories to explain a phenomenon by giving relation between two or more variables. e.g - “okay we need to make it straight so that the force of hitting it makes it big”
10. Task Sentiment (Positive/Negative)	A view of or attitude (emotional valence) toward a situation or event; an overall opinion towards a subject matter. We were interested in looking at positive or negative attitude towards the task that students were working on. e.g - “oh it’s the coolest cage I’ve ever seen, I’d want to be trapped in this cage”, “I’m getting very mad at this cage”
11. Evaluation (Positive/Negative)	Characterization of how a person assesses a previous speaker’s action and problem-solving approach. It can be positive or negative. e.g - ‘oh that’s a pretty good idea”, “no it can’t go like that otherwise it will be stuck”
Non-verbal Behavior (AU - facial action unit)
1. Joy-related	AU 6 (raised lower eyelid) and AU 12 (lip corner puller).	\pbox3cmUsed an open-source software OpenFace for automatic facial landmark detection, and a rule-based approach post-hoc to infer affective states
2. Delight-related	AU 7 (lid tightener) and AU 12 (lip corner puller) and AU 25 (lips part) and AU 26 (jaw drop) and not AU 45 (blink).
3. Surprise-related	AU 1 (inner brow raise) and AU 2 (outer brow raise) and AU 5b (upper lid raise) and AU 26 (jaw drop).
4. Confusion-related	AU 4 (brow lower) and AU 7 (lid tightener) and not AU 12 (lip corner puller).
5. Flow-related	AU 23 (lip tightener) and AU 5 (upper lid raise) and AU 7 (lid tightener) and not AU 15 (lip corner depressor) and not AU 45 (blink) and not AU 2 (outer brow raise).
6. Head Nod	Variance of head pitch.	\pbox3cmUsed OpenFace to extract head orientation, and computed variance post-hoc
7. Head Turn	Variance of head yaw.
8. Lateral Head Inclination	Variance of head roll.
Turn Taking
1. Indegree	A weighted product of number of group members whose turn was responded to (activity) and total time that other people spent on their turn before handing over the floor (silence).	\pbox3cmUsed two novel metrics constructed using an application of social network analysis for weighted data.
2. Outdegree	A weighted product of number of group members to whom floor was given to (participation equality), and the amount of time spent when holding floor before allowing a response (talkativeness).

Table 3. Table 3: Results from semi-automatic verbal behavior annotation. Right column describes external corpus used for training machine learning classifiers & depicts their predictive performance using 10-fold cross validation. Left column depicts inter-rater reliability for human judgment that was used to denoise these behaviors

Verbal Behavior [Krippendorff’s $\alpha$ for human judgment]	Training Data for Semi-Automated Classification [Weighted F1, AUC (10-fold cross validation)]
1. Uncertainty [0.78]	Wikipedia corpus manually annotated for 3122 uncertain 7629 certain instances (Farkas et al., 2010) [0.695, 0.717]
2. Argument [0.792]	Internet Argument Corpus manually annotated for 3079 argument and 2228 non argument instances (Swanson et al., 2015). Argument quality score split at 70% to binarize class label [0.658, 0.706]
3. Justification [process (0.936), causal (0.905), model (0.821), example (0.731), definition (0.78), property (0.847)]	AI2 Elementary Science Questions corpus manually annotated for 6 kinds of justification - process, causal, model, example, definition, property (Jansen et al., 2016). Reported performance is the average performance of 6 binary machine learning classifiers [0.766, 0.696]
4. Suggestion [0.608]	Product reviews (Negi, 2016) and Twitter (Dong et al., 2013) corpuses manually annotated for 1000 explicit suggestion and 13000 explicit non-suggestion instances [0.938, 0.865]
5. Agreement [0.935]	LiveJournal forum and Wikipedia discussion corpuses manually annotated for 2754 agreement and 8905 disagreement instances based on quote and response pairs (Andreas et al., 2012) [0.717, 0.696]

Full text

1 Introduction and Motivation

Curiosity pertains to the strong desire to learn or know more about something or someone, and is an important metacognitive skill to prepare students for lifelong learning [42]. Traditional accounts of curiosity in psychology and neuroscience focus on how it can be evoked via underlying mechanisms such as novelty (features of a stimulus that have not yet been encountered), surprise (violation of expectations), conceptual conflict (existence of multiple incompatible pieces of information), uncertainty (the state of being uncertain), and anticipation of new knowledge ([18, 24]). These knowledge seeking experiences create positive impact on students’ beliefs about their competence in mastering scientific processes, in turn promoting greater breadth and depth of information exploration [43]. These theories have inspired the development of several computer systems aiming to facilitate task performance via enhancing an individual’s curiosity (e.g. [43, 16, 27]), simulating human-like curiosity in autonomous agents [34], and aiding in game theory development [9]. Evoking curiosity in these systems mainly focuses on directing an individual to a specific new knowledge component, followed by facilitating knowledge acquisition through exploration. Such a linear approach largely ignores the how learning is influenced when working in social contexts. Here, a child’s intrinsic motivation, exploratory behaviors, and subsequent learning outcomes may be informed not only by materials available to the child, but also the active work of other children, social and cultural environment, and presence of facilitators [35, 22]. For example, an expression of uncertainty or of a hypothesis about a phenomenon made by one child may cause peers to realize that they too are uncertain about that phenomenon, and therefore initiate working together to overcome the cause of uncertainty, in turn positively impacting their curiosity [20]. While prior literature has extensively studied the intrapersonal origins of curiosity, there seems to be very little prior work on how social factors contribute to moment by moment changes in an individual’s curiosity when learning in social contexts (except for rare exceptions such as [13] that primarily focused on coarse-grained study of adult-child interaction).

As learning in small group becomes prevalent in today’s classrooms [35], it is critical to understand curiosity beyond the individual level to an integrated knowledge-seeking phenomenon shaped by social environment. Embodied Conversational Agents (ECAs) have demonstrated special capacity in supporting learning and collaborative skills for young children [7]. Knowing how social factors influence curiosity allows researchers to design ECAs and other learning technologies to support curiosity-driven learning before children naturally support each other. To address the above goal, we first propose an integrated socio-cognitive account of curiosity based on literature spanning psychology, learning sciences and group dynamics, and empirical observation of an informal learning environment. We make a bipartite distinction between putative functions that contribute to curiosity, and multimodal behaviors that fulfill these functions. These functions comprise (i)“knowledge identification and acquisition” (helps humans realize that there is something they desire to know, and leads to acquisition of the desired new knowledge), and (ii) “knowledge intensification” (escalates the process of knowledge identification or acquisition by providing favorable environment, attitude etc) - at individual and interpersonal level. Second, we perform a statistical validation of this theoretical framework to illuminate predictive relationships between multimodal behaviors, functions (latent variables because they cannot be directly observed) and ground truth curiosity (as judged by naive annotators). A longitudinal latent variable modeling approach called “continuous time structural equation model” [12] is used to explicitly account for group structure and differentiate fine-grained behavioral variations across time.

The main contributions of this work are two-fold: First, it begins to fill the research gap of how social factors, especially interpersonal peer dynamics in group work, influence curiosity. Second, the model is designed to lay a theoretical foundation to inform the design of learning technologies, a virtual peer in the current study, that employ pedagogical strategies to evoke and maintain curiosity in social environments. Findings derived from the current analyses of human-human interaction can be informative in guiding the design of human-agent interaction. Section 2 describes the putative underlying mechanisms of curiosity and associated multimodal behaviors. Section 3 discusses the study context and the annotation approach. Section 4 discusses empirical validation of the theoretical framework of curiosity, with results of the latent variable model fit to our corpus. Section 5 discusses implications and conclusions of our work.

2 Theoretical Framework Development

We initiated development of a theoretical framework for curiosity in learning in social contexts with several iterations of literature review that gradually shifted from individual- to interpersonal-level curiosity. This led us to describe: (i) a set of putative functions that contribute to curiosity, and (ii) multimodal behaviors that provide evidence for potential presence of an individual’s curiosity in the current time-interval because of their fulfillment of these functions.

2.1 Putative Functions that Contribute to Curiosity

The iterative process described above led to emergence of three function groups at the individual and interpersonal level. Each of these functions can be realized in several different behavioral forms. We call the first function group Knowledge Identification. As curiosity arises from a strong desire to obtain new knowledge that is missing or doesn’t match with one’s current beliefs, a critical precondition of this desire is to realize the existence of such knowledge. At an individual level, knowledge identification contributes to curiosity by increasing awareness of gaps in knowledge [29], as well highlighting relationships with related or existing knowledge in order to assimilate new information [8]. Furthermore, exposure to novel and complex stimulus can raise uncertainty, subsequently resulting in conceptual conflict [4, 36]. At an interpersonal level, knowledge identification contributes to curiosity by developing awareness of somebody else in the group having conflicting beliefs [4] and awareness of the knowledge they possess [33], so that a shared conception of the problem can be developed [5].

We call the second function group Knowledge Acquisition. This is because knowledge seeking behaviors driven by curiosity not only contribute to the satisfaction of the initial desire for knowledge, but also potentially lead to further identification of new knowledge. For example, question asking may help close one’s knowledge gap by acquiring desired information from another group member. Depending on the response received, however, it may also lead to escalated uncertainty or conceptual conflict relating to the original question, thus consequently reinforcing curiosity. At an individual level, knowledge acquisition involves finding sensible explanation and new inference for facts that do not agree with existing mental schemata [39, 8], and can be indexed by generation of diverse problem solving approaches [39]. It also comprises comparison with existing knowledge or search for relevant knowledge through external resources to reduce simultaneous opposing beliefs that might stem from the investigation [6]. At an interpersonal level, knowledge acquisition comprises revelation of uncertainties in front of group members [40], joint creation of new interpretations and ideas, engagement in argument to reduce dissonance among peers [19], and critical acceptance of what is told [40].

Finally, we call the third function group Intensification of Knowledge Identification and Acquisition. The intensity of curiosity, or the desire for new knowledge is influenced by factors such as the confidence required to acquire it [29], its incompatibility with existing knowledge, existence of a favorable environment [6] etc. At an individual level, intensification of knowledge identification and acquisition can stem from factors such as anticipation of knowledge discovery [11], interest in the topic [23], willingness to try out tasks beyond ability without fear of failure [21], taking ownership of own learning and being inclined to see knowledge as a product of human inquiry [40]. These factors can subsequently result in a state of increased pleasurable arousal [4]. At an interpersonal level, intensification of knowledge identification and acquisition is influenced by the willingness to get involved in group discussion and the tendency to be part of a cohesive unit [6], and can span from the spectrum of merely continuing interacting to pro-actively reacting to the information others present [5]. Various interpersonal factors play out along different portions of this spectrum. Salient ones include interest in knowing more about a group member [37], promotion of an unconditional positive and non-evaluative regard towards them [11], and awareness of one’s own uncertainty being shared or considered legitimate by those peers [20], all of which can subsequently result in cooperative effort to overcome common blocking points for the group to proceed [11].