Philip Resnik Associate Professor Department of Linguistics and Institute for Advanced Computer Studies People, language, people, computers, people: a.

Philip ResnikAssociate Professor

Department of Linguistics and Institute for Advanced Computer

Studies

People, language, people, computers, people:

a new way for computers to help analysts

New projects

Translation by Iterative Collaboration Between Monolingual Users

Language Evidence for Social Goals: A Linguistic Approach to Persuasion Moves in Discourse

Center for Language and Cultural Analysis

Language Evidence for Social Goals: A Linguistic Approach to Persuasion Moves in Discourse

Center for Language and Cultural Analysis

Why Re-Use Your Towels?

Do it for the environmentCooperate with the hotel and be a partner in

this causeMost guests reuse their towels at least onceThe majority of guests who have stayed in this

room have reused their towels.

12% less effective

18% more effective

33% more effective

Study by social psychologist Robert Cialdini, Arizona State University

Persuasion and Decision ProcessesConstruct ontologies of persuasion tactics

and examine their effectivenessThree theories of interest

Compliance Gaining: taxonomic studiesElaboration Likelihood Model: emphasis on

emotional quality of appealProspect Theory: emphasis on types of

logical appeals

Prospect Theory

Logically, $100 is $100, whether you are winning it or losing it.

But psychologically, $100 lost “counts more” than $100 gained.

Choices are biased by proposal’s framing in terms of gain or loss (Kahneman and Tversky, 1979)

Michael Kinsley's commentary ["Suddenly, health insurers are singing a different tune," Opinion, Aug. 9] elaborates on the insurance lobby's Harry and Louise ads. In contrast to their 1993 edition, the insurers have decided that they will accept all comers if everybody is required to sign up for health-insurance coverage.

All the major variants on reform that are before Congress buy into this. The health-care lobbyists state that "health-care reform is far too importance to be dragged down by divisive political rhetoric."

Sure, there is fine tuning that will have to take place over time; but let's take the first step toward a major overhaul of an inadequate, destructive system that leaves millions without medical care.

Come on, feet draggers, jump on the wagon for decent reform.

• Non-aggressive

• Logical (“central”) route

• Emphasis on gains

Keep up the good work patriots Pinnochios’ strings are comin off! The Liar and Thief is being exposed for the puppet that he is and his puppet masters can’t be happy.

Lookin for more head bashers to beat the people into line. Mount up and arm up Patriots we are in a fight for our country with a bunch of Marxists!

Rally in D.C. Sept 12, 2009 and bring the fight to the chickens door steop!

They must fear for their very existance or they will ram this through anyway, it is the fullfillment of their dreams, bringing America and Capitalism, Free Enterprise to her knees and guaranteeing power eternal for themselves!

• Aggressive

• Emotive gestalt (“peripheral”) route

• Emphasis on risks/losses

Foundational Social Theory

The use of language in persuasionFraming theory (Entman, 1993)

Linguistic choices to “select some aspects of a perceived reality and make them more salient in a communicating text”

Lexical semantics (Dowty 1991, Hopper and Thomson 1980) Connections between surface realizations of

language and underlying construals of events.

Artwork from http://www.amysteinart.com/

“On November 25, a soldier veered his jeep into a crowded market and killed three civilians.”

“On November 25, a soldier’s jeep veered into a crowded market, causing three civilian deaths.”

“The dishwasher broke yesterday.”

“He broke the dishwasher yesterday.”

08/24/09

Language Evidence for Social Goals:A Linguistic Approach to Persuasion Moves in DiscourseUniversity of Maryland (lead), Naval Postgraduate School, Temple University, UC Santa CruzPhilip Resnik (PI), Pranav Anand, Deborah Cai, Craig Martell, Douglas Oard

• Topic• Relate cross-cultural argument styles and language characteristics to persuasion in online media• Automatically detect persuasion attempts in two-party and multi-party dialogue

• Social science perspectives• Persuasion as a decision process• Linguistic characteristics of persuasive interaction• Team includes expert on intercultural communication, persuasion, and negotiation

• Technical approaches• Iterative data acquisition approach in order to characterize observable and latent variables• Bayesian and discriminative modeling to detect persuasion attempts

• Data types• Online text interaction (simulation, bulletin boards, public chat rooms)• Dialogue transcripts

• Languages• English• Arabic (MSA)• Chinese (Mandarin) – starting Option Year 1

• Languages: English, Arabic (MSA), Mandarin

• Social science and language expertise

• Specific online settings and topical domains

Bringing together diverse skills…

Translation by Iterative Collaboration Between Monolingual Users

A real-world problem

International Children’s Digital Library, www.childrenslibrary.orgGoal: 10,000 books translated into 100 languages = 1M book-languages

Wikipedia: 75,000 contributors

Goal:Good and Cheap

MonolingualHumanParticipation

Quality

Aff

orda

bilit

y

MachineTranslation

Professional Bilingual Human Participation

Amateur Bilingual Human Participation

vs. 800 translators

Monolingual Translation Protocol

UI Prototype

Prototype Evaluation

[Submitted to CHI 2010]

=> System seems promising

Sentence fluency

(0=unintelligible, 3=very intelligible)

Sentence accuracy

(1=not translated, 5=full meaning)

MTPrototype

Monolingual Translation Protocol

• Language expertise

• Language expertise

Bringing together diverse skills…

Putting it all together

Bringing together diverse human skills to solve hard

problems

Social science

expertise

Social science

expertise

Knowledge of languages

Knowledge of languages

Understanding of specific topic areas

and domains

Understanding of specific topic areas

and domains

• Human-human interaction (social science)• Human-computer interaction (distributed human computation) • Underlying technology to tackle hard problems (intelligent systems)

• Human-human interaction (social science)• Human-computer interaction (distributed human computation) • Underlying technology to tackle hard problems (intelligent systems)

Understanding of task-

specific goals

Understanding of task-

specific goals