1 of 21

Natural language processing and social interaction

2 of 21

Outline

  1. Motivating problems: why I love these topics, and you should too!
  2. How has NLP+social interaction been affected by this new era (decade, month)?
  3. Alternate course choices
  4. Tentative coursework plans & policies
  5. Individual chats with every student (enrollment issues can be talked about then)

3 of 21

Natural language & social interaction: through the ages!

4 of 21

Natural language & social interaction: through the ages!

5 of 21

Natural language & social interaction: through the ages!

6 of 21

Direct interaction (decision making, persuasion, relationship maintenance, etc.), synchronous (top) or mediated (bottom right): texts, emails, comment threads, etc.

7 of 21

Message spread (and alteration): retweeting, memes, rumors and their effects on people’s opinions

8 of 21

Interactions between groups (political, social, interest-driven sub-communities)

9 of 21

licensed from the Cartoon Bank

younger me

= you

In our own actual lives

I cannot (instantaneously) …� …have better ideas, or� …become alpha dog, or

…be a dog at all.

See many depressing studies on human interpretation and decision making.

10 of 21

licensed from the Cartoon Bank

me

Does choice of phrasing have any impact?

11 of 21

licensed from the Cartoon Bank

Evil me,

trying to disrupt

Influence can be "good" or "bad"

We should understand rhetoric effects —

sometimes in order to counteract them.

12 of 21

"Moderation" can suppress minority views

licensed from the Cartoon Bank

What if I call out that comment?

rule-breaking comment

me�

We should understand moderation's effects.

That doesn't mean it should be used.

13 of 21

Example of the potential influence of language

The framing of an argument emphasizes certain principles or perspectives.

“One of the most important concepts in the study of public opinion” [James Druckman, 2001]

Framing in GMO debates:

"Frankenfood"

"green revolution"

http://www.ourbreathingplanet.com/control-the-world-through-genetically-modified-food/

14 of 21

A sample conversation

  • Wikipedia Article for Deletion discussion ("not a vote"); annotated version (annotations might only display for me)
  • Attempts to persuade in a group setting
  • Conversational derailment into personal attacks
  • Records of changed opinions
  • Lexical innovation (“roadcruft”) (versus think-o’s)
  • Notabilia.net visualization of vote dynamics on selected AfD discussions: dynamics of conversational trends

15 of 21

Why NLP + social interaction?

  • Achieve a better understanding of people�
    • Achieve a better understanding of how to achieve what you want in your life!�
  • Create better systems involving human-human interaction
    • Example: (re)surface conversations someone should be a part of or informed about
    • Not super-emphasized in this class this semesters: building systems that interact with people via natural language

16 of 21

Potential topics (but different papers)

(click through to 2021fa)

This time around: we’ll attempt to focus on work that employs modern NLP techniques.

(Although there is so much good work that isn’t advanced NLP per se!)

(The “interdisciplinary valley”)

Luckily, language modeling has always been a key technology, and now we have better language models!

-> This area has not been “destroyed/solved” by large language models!

17 of 21

Note about style, or, caveats

  • I want to have a very collaborative style in this grad course this semester, so we can all learn together (I am learning with you!).
    • Needs trust, a helpful attitude�
  • I also tend to have a very informal style in my grad courses. �Is this an experience you want to have…?
    • Almost every grad course I’ve taught, my teaching evals have both 5s and 1s. Ones! In a voluntary graduate course!! Don’t waste your time like that!!

18 of 21

Is this the right course for you?

Please take a look at the contents of some of the papers on this quick list of sample papers (URLs should be clickable) before deciding on enrollment; if most of them seem completely impenetrable (or uninteresting), this class may not be the right fit for you.

Related classes: see Cornell's NLP course list.

In particular, Spring 2024 courses CS 6741 Topics in natural language processing and machine learning, CS 5740 Natural language processing (Cornell Tech students only), INFO 4940-LEC 006 Advanced NLP for Humanities Research, CS 4744 (and other crosslists) Computational linguistics I, or CS/IS 4300 Language and information may be a better choice for you; they are excellent courses for sure!

Other classes I am less knowledgeable about: GOVT 3282 Data science applications in political and social research.

The webpage from the last time I (Prof. Lee) taught this class may be useful, as might the webpage from the last time I taught a graduate NLP course.

19 of 21

Tentative plans

  • I’ll do a few initial lectures, and also some foundational material at scheduled intervals
  • Many class meetings will be student presentations of papers

Besides participation, class grades determined as in CS6740 SP23:

  • Midterm paper that reviews and critically analyzes the class material — tell me what you’ve learned so far.
  • Final paper that reviews and critically analyzes the class material — tell me what you’ve learned
  • To receive an A: in your presentations, demonstrate active intellectual engagement with the papers you are required to present, beyond simply summarizing them; participate meaningfully in at least 85% of the discussions; in the midterm paper, demonstrate active intellectual engagement with the material in the course so far, including thoughtful synthesis of common trends and issues across multiple papers; in the final paper, do so across at least two subfields.

No course project: I expect everyone’s got their own research to focus on in “real life”.

20 of 21

Meeting schedule from LAST YEAR, to get a feeling of the rhythm.

Readers: aim to have questions highlighted ~1.5 days before presentation date.

Lec 12/M Mar 13

YG leading “constitutional AI” [notes/links]. Readers: RH, BW

Lec 13/W Mar 15

VP leading “automatic chain of thought”. [notes/links] Reader: YG

YS for “how many data points is a prompt worth?” [notes/links]. Reader: -

Lec 14/M Mar 20

BW leading “probing” [notes/links]. Reader: YSTY for “Does BERT rediscover…?” [notes/links]. Reader: VP

Lec 15/W Mar 22

RH leading “adapting for historical langs” [notes/links]. Reader: TYYG for “semantic shift/mental health” [notes/links]. Reader: BW

Lec 16/W Mar 29

[then spring break!]

YS leading “discriminators, not generators” [notes/links]. YG

VP for “prompt tuning for discr” [notes/links]. Reader: RH

Lec 17/W Apr 12

TY leading “Flamingo” [notes/links]. Reader VP

RH for leading “CoAuthor” [notes/links]. Reader: YS

Lec 18/M Apr 17

BW leading “S4” [notes/links]. Reader: TY

21 of 21

Next: in-person individual get-to-know-you “interviews”

The “interview” questions:

  1. How does this class fit your overall research/educational goals?
  2. What is your background in NLP, machine learning, artificial intelligence, etc.
  3. Enrollment issues?