iiinova — where your digital self keeps evolving

Khuyen Le @knle.bsky.social · 16/07/2026

But many questions remain: How do children represent speaker knowledge states? What are the temporal dynamics of the cue integration process? How do these representations / processes change across development? We’re excited to continue exploring these questions! 9/9

020

Khuyen Le @knle.bsky.social · 16/07/2026

Our findings suggest that learners don't simply accumulate word-object statistics to learn words, but use speaker knowledge to constrain cross-situational learning. We're now exploring how children integrate implicit, context-specific epistemic cues with statistical cues during word learning. 8/

120

Khuyen Le @knle.bsky.social · 16/07/2026

We also measured children's Theory of Mind using a battery of classic tasks (from Wellman & Liu), but performance on this measure didn't explain individual differences in word learning. We discuss a few possible reasons for this in the paper. 7/

120

Khuyen Le @knle.bsky.social · 16/07/2026

Crucially, the "unknown" obj. was the higher-probability foil based on co-occurrence statistics. But when ppts make mistakes, they chose the "unknown" obj. less often in the Ignorance condition than in the 3-Object Control. They also chose this object less than chance in the Ignorance condition. 6/

120

Khuyen Le @knle.bsky.social · 16/07/2026

If they rely only on co-occurrence statistics, performance in the Ignorance condition should resemble the 3-Object Control. Our results supported the first hypothesis: Performance in the Ignorance condition was better than in the 3-Object Control and no different from the 2-Object Control. 5/

120

Khuyen Le @knle.bsky.social · 16/07/2026

If learners use the speaker's ignorance to rule out the "unknown" object, they should learn words better in the Ignorance condition compared to 3-Object Control, and similar to the 2-Object Control. 4/

110

Khuyen Le @knle.bsky.social · 16/07/2026

On each learning trial in the Ignorance condition, ppts heard 2 novel words while viewing 3 objects. The speaker then pointed to one object and said she didn't know what it was called. Two control groups completed identical tasks with either 2 or 3 objects, without the ignorance statement. 3/

121

Khuyen Le @knle.bsky.social · 16/07/2026

The debate: do learners track all word-object co-occurrences cross-situationally to learn words, or do they use information about a speaker's mental states to narrow down the set of possible meanings? Children (3.5-6yos) and adults completed a CSWL task, with speaker knowledge manipulation. 2/

120

Khuyen Le @knle.bsky.social · 16/07/2026

How do children learn new words? Do they simply track word-object co-occurrences across contexts, or do they also use information about what speakers know and intend? With @mzettersten.bsky.social and @drbarner.bsky.social, I tackled this question in our new preprint! 1/🧵👇 osf.io/preprints/ps...

osf.io

OSF

1244

Khuyen Le @knle.bsky.social · 06/06/2026

If you're curious, an interactive example of the digital Give-N task is here! khuyen-le.github.io/give-n-onlin... (no data is collected, though that might be my summer pet project)

011

Khuyen Le @knle.bsky.social · 06/06/2026

Takeaway: the digital Give-N task is a useful and practical alternative for measuring knower levels! Though we recommend additional validation when the task is modified and used to compare against results from a physical version, or in communities with low access to screen-based tech. 4/4

110

Khuyen Le @knle.bsky.social · 06/06/2026

We also replicated some previous findings re: knower level consistency in Give-N: - Low agreement for 3-, 4-, 5-knowers - No effect of task order One caveat is that children performed better (on a trial basis) on the physical version, with minimal impact to knower level classification. 3/

120

Khuyen Le @knle.bsky.social · 06/06/2026

We tested kids on both physical and digital versions of Give-N and found consistent knower level assignments across versions, regardless of item arrangement (below), and with multiple agreement analyses (e.g., individual knower levels, tripartite classifications of non-, subset-, and CP-knowers). 2/

121

Khuyen Le @knle.bsky.social · 06/06/2026

If you study number and want a more standardized and convenient version for Give-N, consider using a digital task! New preprint 🚨 (with Kenyee Liu, Daniel Hyde and @drbarner.bsky.social): a digital Give-N task shows high knower level agreement with a physical version. 1/ osf.io/preprints/ps...

171

Reposted by Khuyen Le

Sophie Arnold @sophieharnold.bsky.social · 04/06/2026

New paper with @mohitmukherji.bsky.social, Mona Lebrun, and @marjorierhodes.bsky.social @JEP:G! Did you ever wonder whether saying things like “boys can wear dresses too” is effective at reducing stereotypical inferences? We find that they’re probably not—check out more here: doi.org/10.1037/xge0...

doi.org

APA PsycNet

3209

Khuyen Le @knle.bsky.social · 30/05/2026

Kids compute scalar implicatures at a younger age than previously found, when they can form their own interpretation first before we present them with irrelevant alternatives! Another case of task pragmatics masking kids’ early competence 🤔

020

Khuyen Le @knle.bsky.social · 29/05/2026

Why do some nouns have different mass/count statuses in different languages? We made some proposals in the paper, but there's much more to explore!

010

Khuyen Le @knle.bsky.social · 29/05/2026

Overall, our findings support the view that OMNs share the same individuation semantics as count nouns. That still leaves lots of open questions, though: For example, how do children acquire OMNs if their semantics overlap with count nouns? Why are some collective nouns mass and others count?

120

Khuyen Le @knle.bsky.social · 29/05/2026

... which appear in count syntax (e.g., English "furniture" vs. French "meubles"). Again, we found no difference: both English OMNs and their French count noun counterparts were similarly influenced by context. And both English and French speakers generally quantified them over individuals.

110

Khuyen Le @knle.bsky.social · 29/05/2026

One concern is that differences in lexical meaning between individual nouns could mask the effect of syntax on quantification (this is also a problem for earlier work on this topic). So we compared judgments of English OMNs with French speakers' judgment of their French translations...

110

Khuyen Le @knle.bsky.social · 29/05/2026

And... we found that OMNs were quantified overwhelmingly by number, much like count nouns. And importantly, they were no more sensitive to contextual manipulation than count nouns were!

120

Khuyen Le @knle.bsky.social · 29/05/2026

We collected quantity judgments in a neutral context and in a context emphasizing functional fulfillment, a dimension previously theorized to be part of the semantics of OMNs. (A separate study confirmed that participants judged the more varied group as better able to fulfill the relevant function.)

110

Khuyen Le @knle.bsky.social · 29/05/2026

We compared English speakers’ quantity judgments for nouns like "furniture" (object-mass nouns, OMNs) to collective count nouns like "tools." Participants compared groups that traded off number vs. variety: one group had fewer items but more variety, while the other had more items but less variety.

110

Khuyen Le @knle.bsky.social · 29/05/2026

Do they have the same individuation semantics as count nouns? Or are they semantically underdetermined, allowing different contexts to flexibly determine how they're quantified?

110

Khuyen Le @knle.bsky.social · 29/05/2026

Languages like English distinguish between mass nouns (e.g., water) and count nouns (e.g., cat). One way these nouns differ that count nouns quantify over individuals, while mass nouns supposedly do not. But what about mass nouns like "furniture", which often seem to quantify by number?

181

Reposted by Khuyen Le

Martin Zettersten @mzettersten.bsky.social · 27/02/2026

Our department is hiring an Assistant Teaching Professor!! This is a joint-appointed position with Computational Social Sciences (css.ucsd.edu). It's 75+ degrees F and sunny today, just thought I'd mention apol-recruit.ucsd.edu/JPF04461

apol-recruit.ucsd.edu

Assistant Teaching Professor in Computational Social Science and Cognitive Science

University of California, San Diego is hiring. Apply now!

14428

Khuyen Le @knle.bsky.social · 27/02/2026

So, older preschoolers use speaker-specific epistemic reasoning to make mutual exclusivity inferences and interpret the meanings of novel labels. Though other assumptions, such as linguistic conventionality of labels, are no doubt important as well! (7/7)

000

Khuyen Le @knle.bsky.social · 27/02/2026

We also tested children’s general Theory of Mind abilities and found that this predicts their ability to make adult-like evaluations of epistemic knowledge. Many open questions about how children draw on ToM abilities to reason about others’ linguistic knowledge and learn words! (6/7)

100

Khuyen Le @knle.bsky.social · 27/02/2026

We found that from 4.5yos, children made more mutual exclusivity inferences when the label was taught compared to when it was invented. Children also made more mutual exclusivity inferences when they believed the speaker knew the introduced label, regardless of how it was introduced. (5/7)

100

Khuyen Le @knle.bsky.social · 27/02/2026

We asked children whether the absent speaker knew this label. After the absent speaker returned and made a request using another word (e.g., ‘bem’), we asked whether children would exclude the already-labeled object from consideration. (4/7)

100

Khuyen Le @knle.bsky.social · 27/02/2026

We expanded previous paradigms to differentiate these possibilities, by manipulating whether a label (e.g., ‘dax’) is taught vs. invented in the absence of a speaker (see Figure!) (3/7)

100

Khuyen Le @knle.bsky.social · 27/02/2026

When children make mutual exclusivity inferences, they can do so by reasoning about what other labels a speaker knows (a la Gricean inferences), projecting their own knowledge onto the speaker, or reasoning generally that labels are shared by all speakers of the same language. (2/7)

100

Khuyen Le @knle.bsky.social · 27/02/2026

Check out my new paper with @drbarner.bsky.social in JECP! We asked whether mutual exclusivity inferences involve epistemic reasoning about what a speaker knows, and whether children can infer speakers' knowledge of words from linguistic conventionality. (1/7) www.sciencedirect.com/science/arti...

sciencedirect.com

The role of epistemic reasoning in mutual exclusivity inferences

When encountering a novel word, adults and children as young as 12 months old often reason that it refers to a novel object rather than one with an ex…

1196

Reposted by Khuyen Le

samuel mehr @mehr.nz · 14/12/2023

it's wild that R, the ubiquitous statistical computing language, was co-created by a Māori prof (Ross Ihaka) — and yet the vast majority of scientists who use R don't know this is like inventing the toaster. possibly the largest impact of a single member of an indigenous community on modern science

441593628

Reposted by Khuyen Le

Josh de Leeuw @joshdeleeuw.bsky.social · 19/09/2024

We are currently conducting interviews with jsPsych users to help us shape long-term project goals. We are interested in speaking with folks with all levels of comfort with jsPsych. The interviews are happening over the next four weeks. Each session is 20-30 minutes. We are paying participants $20.

244

Khuyen Le @knle.bsky.social · 21/05/2024

New preprint by the phenomenal @ebruevcen.bsky.social with @drbarner.bsky.social revealing that people interpret conditionals pragmatically, initially treating them as biconditionals! So excited to see where this work goes next, especially for children’s acquisition of conditionals!

040

Reposted by Khuyen Le

tj mahr 🤘 @tjmahr.com · 03/05/2024

notes on citing R and R Packages #rstats www.tjmahr.com/r-package-ci...

tjmahr.com

Notes on Citing R and R Packages

Who, what, where, when and which

02013

Reposted by Khuyen Le

David Barner @drbarner.bsky.social · 01/05/2024

Another paper w/ @urvi.bsky.social, showing that Hindi kids learn yesterday/tomorrow earlier than English kids, despite Hindi expressing these with just one word 'kal'. We argue this is evidence for the priority of tense information (over associations with events). osf.io/preprints/os...

osf.io

OSF

031

Reposted by Khuyen Le

David Barner @drbarner.bsky.social · 01/05/2024

Happy to share this paper led by the fabulous @urvi.bsky.social. Turns out that children's struggle to understand temporal language may be partly because the things we refer to in tests of knowledge are not actually in the past or future & rely on hypothetical reasoning about imaginary timelines.

0133

Khuyen Le @knle.bsky.social · 02/05/2024

This suggests that reasoning about how number words encode exact quantities and how numbers relate to exactly equal sets emerges after learning to count. Exciting questions remain about how such learning happens! (4/4)

010

Khuyen Le @knle.bsky.social · 02/05/2024

But children who can make this inference might not do so based on exact equality: they sometimes assign the same number word to two sets that are approximately equal, but differ by just 1 item❗(3/4)

111

Khuyen Le @knle.bsky.social · 02/05/2024

Children who can count and construct large sets (CP-knowers in the literature) can infer the quantity of a hidden set from a set that appears equal, but children who only know small number words (subset-knowers) cannot 🧒 (2/4)

110

Khuyen Le @knle.bsky.social · 02/05/2024

New preprint with Rose Schneider & @drbarner.bsky.social on children's learning of number concepts! osf.io/preprints/ps... TL;DR: Learning to count large sets may be necessary, but not enough for understanding that numbers represent exact quantities 🧠🔢 (1/4)

osf.io

OSF

172