Carolyn Penstein Ros 1 Theoretical framework Psychology-> - - PowerPoint PPT Presentation

Carolyn Penstein Rosé

1

• Theoretical framework
• Psychology-> Sociolinguistics -> Language Technologies
• Authoritativeness: Vertical power distance
• Results using Integer Linear Programming
• Transactivity: Horizontal power distance
• Results using Support Vector Machines and Dynamic

Bayesian Networks

• Applications in Learning Technologies and

Health Informatics

2

3

• Basic concepts of power and

social distance explain social processes operating in interactions

• Social processes are reflected

through patterns of language variation

• If we understand this connection,

we can model language more effectively

• Models that embody these

structures will be able to predict social processes from interaction data Psychology Sociolinguistics Language Technologies

4

• Some influential theories
• Social Identity Theory

(Brewer, 1997)

• Self-Categorization

Theory (Turner, 1985)

• Social Cognitive Theory

(Bandura, 1986)

• We gain influence in

interaction through manipulation of horizontal and vertical social distance

• We manipulate distance

through signaling

Person Authority Identity

Vertical Social Distance Horizontal Social Distance

A2

Instructing, suggesting, or requesting non-verbal action

A1

Narrating or performing your

• wn non-verbal action

5

Additionally… ch (direct challenge to previous utterance)

• (all other moves, backchannels, etc.)

K2

requesting information, knowledge, opinions, or facts

K1

giving opinions, knowledge, information, or facts

Authoritativeness: # Source Core Moves # Core Moves

Constraints for Integer Linear Programming…

(Martin and Rose, 2003)

6

1.You don’t request information or action after it’s been given. 2.Knowledge and action don’t mix. 3.You don’t respond to the same request twice. 4.You don’t respond to your own requests.

Machine Learning for Negotiation

Data: 20 hand-coded conversations (4374 turns) Results given are from 20-fold leave-one- conversation-out cross validation All improvements between models are significant (p < .01) Tools used:

• SIDE (Mayfield and Rosé, 2010) for feature extraction
• SVMlight (Joachims, 1999) for machine learning
• Learning-Based Java (Rizzolo and Roth, 2010) for ILP inference

7

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Kappa Accuracy Authoritativeness r^2 Baseline Baseline+ILP Contextual Contextual+ILP Human (Upper Bound)

8

0.58 0.68 0.95

In pairs, ratio between students’ Authoritativeness ratios predicted group self-efficacy. More authoritative students also showed more warning signs of aggressive (“bullying”) behavior. In MapTask data, groups with more authoritative instruction givers produce more errors.

9

Is Authority useful? Yes!

(joint work with Iris Howley, CSCL 2011)

10

• Explicitly display

reasoning

• Orient contributions

towards previous contributions

• [Student1] Well…. U do

know that increasing tmax and pmax means more Qin

• [Student2] yeah, that

makes an argument for not using that idea in our design – but on the other side, it leads to more quality – which means you get more work out of the turbine

Evaluating Context-Based Features

0.52 0.71 0.61 0.67 0.73 0.62 0.66 0.70 0.61 0.69 0.73 0.61 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 Social Macro Micro

Dimension Kappa from 10-fold CV

Base Base+Thread Base+Seq Base+AllContext

Baseline Base + Thread

Features

unigram, bigram, length depth, parent_child_similarity fsm

12

• Modeling speech style

accommodation using dynamic Bayesian networks (Jain et al., submitted)

• Leveraging the idea that

social processes are continuous rather than instantaneous

• Correlating speech style

accommodation with transactivity (Gweon et al.,

submitted)

R2 = 0.18

• Both transactivity and authoritativeness

correlate with learning

• Applications in online assessment of group

learning

• Triggering context sensitive support for group

learning

• Recent demonstration of generalization to

doctor-patient interactions

• Authoritativeness predicts some important patient

perception metrics

• Automatic analysis may support doctor

professional development

13

Thank You! Questions?

14

Do you have a camera shop? Yeah. Okay, first go down along the left hand side of that. How far down do I go?

K2 K1 A2 K2

All the way down to the bottom left corner. K1

A1

Right.

15

Right.

Now go up to the left

• f the youth hostel.

Where’s that? Right above the alpine garden. I don’t have that.

A2 K2 K1 ch

16