Social and political psychologist. NYU professor. Music appreciator. Californian in Manhattan.
Reposted by Eric D. Knowles
Reposted by Eric D. Knowles
This is the United States of America. We shouldn't be afraid of our government — our government should be afraid of us.
Reposted by Linda R. Tropp
Reposted by Timothy Snyder, Robert West, Rebecca Solnit , and 34 more Timothy Snyder, Robert West, Rebecca Solnit, Dorothy Bishop, Will Jennings, Samuel L. Perry, Tom Shakespeare, Valerie P. Hans, Henning Melber, Philip Nel, Peter Holmes, Joni Lovenduski, Trevon D. Logan, Robert C. Richards, R. M. Wilson, Gerald Friedman, Nazita Lajevardi, Scott A. Imberman, Michael H. Whitworth, Anne Norton, Eric D. Knowles, Michael Müller, Helen Jefferson Lenskyj, Daryl Adair, Mark Priestley, Kate Starbird, Pauline Stafford, Björn Heile, Umut Özkırımlı, Greg Linden, Tim Stephens, Christina Pagel, Colin Talbot, Markus Haverland, Jeannette Sutton, Virginia Sapiro, Jim McManus
www.startribune.com/she-was-an-a...
Reposted by Eric D. Knowles
Reposted by Elizabeth Saunders, Steven Heydemann, Norbert Schwarz , and 13 more Elizabeth Saunders, Steven Heydemann, Norbert Schwarz, Michael A. Clemens, Michelle Everson, Robert C. Richards, McKenzie Wark, Roopika Risam, Beatriz Gallardo Paúls, Samantha Brennan, Michael D. McDonald, Anne Norton, Eric D. Knowles, Mark Rice, Olúfẹ́mi Táíwò, David Darmofal
www.mprnews.org/story/2026/0...
Yes, I know it was created by an act of congress. So do to it what Trump did to USAID and the Education Department. Starve it out of any meaningful existence. Bury it.
Reposted by Robb Willer, Eric D. Knowles
donmoynihan.substack.com/p/when-volat...
lexichron is a work-in-progress but stable for many use cases. Check out the README and included Jupyter notebooks for more info..
I welcome your feedback!
Analysis resources:
- Easily compute cosine similarities and plot temporal trends
- Track and visualize semantic drift for individual words
- Compute WEATs
Training and evaluation pipelines:
- Hyperparameter tuning: architecture (SGNS vs. CBOW), vector size, context window, weighting strategy, etc.
- Model evaluation using similarity and analogy benchmarks
- Visualizations comparing model quality
- Regressions quantifying hyperparameter impact
Purpose-built for HPC clusters and cloud-computing. Pipelines that take weeks on a laptop run in minutes or hours on a cluster. (Fewer cores? It'll work, just slower.)
lexichron truly shines with 30+ CPUs, 80+ GB RAM, and fast NVMe SSD storage.
Key features:
- Tunable preprocessing, including case normalization, lemmatization, stop-word removal, and spell-checking
- Vocabulary whitelisting for efficient filtering
- Bigram preservation, allowing retention of semantically interesting word pairs
- RocksDB databases for fast queries
Both data-prep pipelines feed into a single word2vec training module (gensim's implementation) for training, normalizing, and aligning yearly models to capture semantic change.
lexichron implements two data-prep pipelines:
- Google Ngrams: download and filter 1–5 grams in 8 languages
- Mark Davies' corpora: process datasets from English-Corpora.org containing year and genre metadata (e.g., COHA, COCA; this requires a license and access to the corpus files.)
lexichron is for tracking conceptual shifts across years and decades. It efficiently prepares, filters, trains models on, and analyzes results from extremely large corpora.
Although it can be tuned for personal computers, lexichron is designed for parallelization on HPCs and cloud platforms.
Reposted by Robb Willer, Eric Hehman
I've created a toolkit — lexichron — for measuring long-term shifts in word meanings using Google Ngrams and other corpora (e.g., COHA, COCA).
Here's the public GitHub repo:
github.com/eric-d-knowl...
🧵
“.. we’re not even 24 hours into this .. Decades later, we’re still dealing with the aftermath in Afghanistan and Iraq.”
fallows.substack.com/p/blind-into...
“.. we’re not even 24 hours into this .. Decades later, we’re still dealing with the aftermath in Afghanistan and Iraq.”
fallows.substack.com/p/blind-into...
Reposted by Jennifer A. Richeson, David R. Miller, Eric D. Knowles