iiinova — where your digital self keeps evolving

naitian

@naitian.org

1K followers 873 following 184 posts

NLP / CSS PhD at Berkeley I School. I develop computational methods to study culture as a social language.

Posts Replies Media

naitian @naitian.org · 05/02/2026

ew why is claude doing that

Claude Code says that it has been Canoodling for 2 minutes and 46 seconds.

000

naitian @naitian.org · 07/01/2026

I just got back from Michigan and—

Screenshot of Apple weather app for Berkeley, CA with a warning for an “extreme cold watch”. Today shows currently mostly sunny 53 degrees with a high of 57 and low of 48.

naitian @naitian.org · 05/12/2025

We also provide dialogue act annotations and timestamps of when players completed subtasks for each level. We can use these annotations to show that, e.g., directives are more common before the completion of hard tasks, while confirmations are more common after. 3/n

Figure from the paper which shows how, for difficult tasks, directives and options tend to increase immediately before tasks are completed and drop afterwards, whereas the inverse is true for confirmations and expressives.

naitian @naitian.org · 05/12/2025

Many linguistic phenomena like spatial reference or convention formation are underrepresented in existing datasets, because they’re not embodied in complex or goal-oriented environments. In this clip, players combine both language and movement to successfully establish left vs right. 2/n

naitian @naitian.org · 05/12/2025

A couple years (!) in the making: we’re releasing a new corpus of embodied, collaborative problem solving dialogues. We paid 36 people to play Portal 2’s co-op mode and collected their speech + game recordings. Paper: arxiv.org/abs/2512.03381 Website: berkeley-nlp.github.io/portal-dialo... 1/n

A figure demonstrating the different aspects of the corpus described in the tweet. There is a main isomorphic 3D view of a level in the Portal 2 co-op game, with some portals, lasers, and the blue and orange players. Inset, there are first-person captures of the blue and orange player views. There is also a box containing the transcribed dialogue with timestamps and labels for the discursive acts. Finally, there is a box containing a task and a list of subtasks. Some subtasks are already crossed out, with the time that they have been completed. The last subtask ("Player 2 places portal 4 on wall 4") is marked incomplete.

The dialogue is as follows:

Blue: Can you put your other portal up here? (tagged as directive)
Orange: Where? (tagged as request for clarification)
Blue: On uh, on this wall. (tagged as directive)
Blue: So that it uh points at the circle. (tagged as directive)
Orange: Okay. (tagged as commit)

The full list of subtasks is:

Task: Redirect lasers
Subtask: Player 1 places portal 1 on wall 1. (completed)
Subtask: Player 1 polaces portal 2 on wall 2 or 3. (completed)
Subtask: Player 2 places portal 3 opposite of portal 2. (completed)
Subtask: Player 2 places portal 4 on wall 4. (incomplete)

naitian @naitian.org · 20/09/2025

Centralisation of the means of production and socialisation of labour at last reach a point where they become incompatible with their capitalist integument. This integument is burst asunder. The knell of capitalist private property sounds. The expropriators are expropriated.
The capitalist mode of appropriation, the result of the capitalist mode of production, produces capitalist private property. This is the first negation of individual private property, as founded on the labour of the proprietor. But capitalist production begets, with the inexorability of a law of Nature, its own negation. It is the negation of negation. This does not re-establish private property for the producer, but gives him individual property based on the acquisition of the capitalist era: i.e., on cooperation and the possession in common of the land and of the means of production.

naitian @naitian.org · 23/07/2025

Speaking of #ic2s2, @amirgo.bsky.social's keynote talks about cultural differences are not necessarily oppositional, and we should consider other kinds of relations --- an idea that's also in sociocultural linguistics, and which we mention in the paper!

Amir Goldberg on stage at IC2S2. The slide behind him highlights the difference between orthogonality (are they different but not opposed?) and oppositionality (are they opposed?) in evaluating the similarity of interpretations.

A screenshot from "Culture is not Trivia":

Similarity and difference are not the only relations available within this framework, which also includes authentication-denaturalization
and authorization-illegitimization, among others. If the cultural space is structured through these relations, computational methods might benefit from considering how to encode them.

The first phrase is highlighted.

naitian @naitian.org · 23/07/2025

I'm thrilled to be doing an oral presentation on "Culture is not Trivia" at #ACL2025 next Wednesday 7/30, as well as participating in the human-centered NLP panel afterwards! (thanks also @lauraknelson.bsky.social for the shoutout in her #ic2s2 keynote today!) aclanthology.org/2025.acl-lon...

A poster for "Culture is not Trivia: sociocultural theory for cultural NLP" which takes the form of a flow-chart. The central question, and the starting point of the flow chart, is "What is culture in cultural NLP?"

An arrow is labeled "wait, so what's cultural NLP?" This leads to a block explaining that the goals of cultural NLP are described in section 2 of the paper. They include inclusivity, depth, discerning, and adaptiveness.

That leads to an arrow that says "that sounds great!". But there are recurring challenges in this kind of work! Section 3 surveys some of these: a discomfort around the proxies being chosen, a lack of coverage, and a lack of dynamicity.

That in turn leads to an arrow labeled "Hm, sounds like we need to figure out..." and it leads back to the main question: "What is culture in cultural NLP?"

A final arrow extends below this block: "Well, who's to say, really?"

This points to sociocultural linguistics. Section 4 explores how other disciplines, like sociolinguistics, linguistic anthropology, and discourse analysis have faced similar challenges in the past. Section 4.2 gives an overview of sociocultural linguistics, which is a set of principles tying together some convergent themes: emergence, positionality, indexicality, relationality, and partialness.

One arrow extends from this asking, "what's that have to do with cultural NLP?" Section 5 gives a case study of how indexicality clarifies how to think about stereotypies in the context of mining cultual knowledge from the web.

Another arrow says "How can I build safe NLP systems?" Section 6.2 explores how localization can serve as a useful model from building culturally aware technologies because it forces developers to define culture explicitly and tractably.

Finally, an arrow asks "how can I study culture with NLP methods?" Section 6.1 lays out theoretically motivated directions for future empirical and theoretical work in computationally modeling culture.

naitian @naitian.org · 02/07/2025

Still image from season two of Nathan Fielder's show "The Rehearsal", depicting a large hall meant to evoke imagery of Nazi Germany, with large Paramount banners hanging in the background.

000

naitian @naitian.org · 19/05/2025

And I think lost poetry provides a counterexample to this claim:

A quote from the original paper: The elusiveness makes sense for money and politics, but up to now, it has not been reasonable - or even considered possible -- for art. GenAI images make it possible.

naitian @naitian.org · 10/03/2025

go bears!

A screenshot of the terminal showing model training progress. The label for one of the progress bars says "Calidation Dataloader" instead of "Validation Dataloader".

naitian @naitian.org · 18/02/2025

There's been a lot of work on "culture" in NLP, but not much agreement on what it is. A position paper by me, @dbamman.bsky.social, and @ibleaman.bsky.social on cultural NLP: what we want, what we have, and how sociocultural linguistics can clarify things. Website: naitian.org/culture-not-... 1/n

Culture is not trivia: sociocultural theory for cultural NLP. By Naitian Zhou and David Bamman from the Berkeley School of Information and Isaac L. Bleaman from Berkeley Linguistics.

naitian @naitian.org · 16/02/2025

Ralph Wiggum of the Simpsons sitting on a bus: “I’m in danger,” he says while chuckling.

naitian @naitian.org · 16/12/2024

I've been obsessed with Sotheby's auction videos bc there's such an internally coherent set of language and gesture that the auctioneers use and because there is one pose they all do that reminds me of pointer dogs

Sotheby's auctioneer Oliver Barker eliciting bids, living up to his surname and leaning forward on the rostrum in a pose that evokes the image of a bird dog at alert.

A very good bird dog, who is posed similarly to Sotheby's auctioneer Oliver Barker.

Sotheby's auctioneer Phyllis Kao leaning forward on the rostrum and pointing behind her, evoking the form of a pointer dog with tail pointing straight backwards.

A spotted pointer dog with tail outstretched behind him, in a way that very much resembles esteemed Sotheby's auctioneer Phyllis Kao.

naitian @naitian.org · 12/12/2024

I love reading conversation analysis papers bc sometimes it just feels like watching TV

An excerpt from a conversation analysis paper describing a scene during a journalism hackathon where some group members (Kat, Mia, and Jen) are unsure whether a fourth member (Lee) is joining their group, so Kat semi-awkwardly (though this could just be a projection of my own social anxiety) broaches the subject directly.

Illustrations next to the transcript describe Kat waving at Lee, and Kat and Jen celebrating after Lee affirms his participation.

naitian @naitian.org · 19/11/2024

Emotional range reflects discursive patterns: phrases with low range tend to be grounded in restrictive interactional settings, while those with high range are more open-ended. Performance on screen draws on the same semiotic toolbox that the audience interacts with day-to-day! 5/n

Table showing some phrases with high and low emotional range, and their corresponding entropy values. The phrase with the lowest emotional range is “Could I ask you something?” and the phrase with the highest emotional range is “All rise.”

naitian @naitian.org · 19/11/2024

We use this parallel dataset of text and audio to answer questions about narrative trajectory and emotionality. We find that emotionality in general has decreased over the years, even after controlling for the dialogue. This means the effect isn’t just due to a shift in writing over the years. 3/n

Line graph from the paper that shows the average emotionality of films has decreased since 1980, reaching its lowest point around 2010.

naitian @naitian.org · 19/11/2024

Taking advantage of our large corpus of contemporary American film, we create a pipeline of speech emotion recognition and transcription models to match the words being spoken with information about how they were delivered. 2/n

Diagram showing three concepts from the paper: First, dialogue phrase groups, which are groups of semantically equivalent phrases generated by clustering audio transcriptions. Second, utterances, which are pieces of audio extracted using speech segmentation. Finally, emotion labels for each utterance are classified with a speech emotion recognition model. Multiple utterances are associated with each dialogue phrase group in a many-to-one relationship.

naitian @naitian.org · 19/11/2024

🎬 Coming soon to a theater near you!🍿 Film is a semiotically rich medium: meaning is conveyed through the music, visuals, language, and more. A new paper from me and @dbamman.bsky.social explores what it means to computationally study performance in film. Website: naitian.org/once-more-wi... 1/n

Screenshot of the abstract for the paper, "Once More, With Feeling: Measuring Emotion of Acting Performances in Contemporary American Film". The abstract reads: "Narrative film is a composition of writing, cinematography, editing, and performance. While much computational work has focused on the writing or visual style in film, we conduct in this paper a computational exploration of acting performance. Applying speech emotion recognition modes and a variationist sociolinguistic analytical framework to a corpus of popular, contemporary American film, we find narrative structure, diachronic shifts, and genre- and dialogue-based constraints located in spoken performances.

naitian @naitian.org · 06/11/2024

election anxiety is grabbing the raw data to calculate my own stats

screenshot of a jupyter notebook showing some code fetching vote count data from the washington post

naitian @naitian.org · 09/07/2024

Check out my library card collection

A pile of library cards collected from all over the world

naitian @naitian.org · 21/04/2024

I'm already 10x-ing my writing with AI.

Screenshot from Google Docs where the autocompletion suggests I want to write about a "disciplinary discipline".

naitian @naitian.org · 16/11/2023

Doing all this required a huge engineering effort in which we developed a pipeline to process over 27.9M images posted to Reddit between 2011-2021. We make the semantic clusters publicly available, as well as URLs to fetch all the images.

naitian @naitian.org · 16/11/2023

We use these clusters to surface subreddits that use visually different templates to convey the same semantics. We find not only do subreddits prefer certain styles over others, but they choose ones that index localized cultural knowledge!

naitian @naitian.org · 16/11/2023

Recalling distributional semantics, we fine-tune an LM to learn how fill text aligns with the template, giving semantic embeddings for templates that we can cluster. Visually diverse clusters emerge even for complex semantic functions!

naitian @naitian.org · 16/11/2023

We take advantage of the unique compositional multimodality of memes to learn the semantics of meme templates without supervision by breaking them down into the base image template and the text fill that goes inside.

naitian @naitian.org · 16/11/2023

Memes are pervasive in online speech. Do they have the socially meaningful variation we see in other aspects of language? YES! Preprint from me, @davidjurgens.bsky.social and @dbamman.bsky.social on the semantic structure and visual diversity of 3.8M Reddit memes. naitian.org/social-memeing

naitian @naitian.org · 02/10/2023

is this... interdisciplinarity?

The listings for two papers about shibboleths in Google scholar, one of which is about enregisterment and identity in Michigan's Keweenaw Peninsula, and the other is about SAML and resource access management standards.