Martin Ruskov @mapto@qoto.org

Today at #CHR2025, I will be presenting our work on the evaluation of the historical adequacy of masked language models (MLMs) for #Latin. There are several models like this, and they represent the current state of the art for a number of downstream tasks, like semantic change and text reuse detection. However, a historical researcher, philologist or else would want to be sure that such models really represent the historical period of interest. For example, it would be an embarrasing hallucination if St. Augustine showed up in the context of the Roman senate.

Our evaluation confirms a known problem: LLMs and masked models in particular are trained on corpora without attention to historical periods. Unlike other research we've done on Early Modern English, this problem leads to models being barely distinguishable when it comes to their ability to generate based on a historical period. Even though history is a case where it is most obvious when models go wrong, this type of contamination is a known problem for LLM training overall, think of different legal jurisdictions using the same language, dialects in programming languages, etc.

This research was generously supported by AgileLab.

The full paper is available at:
https://anthology.ach.org/volumes/vol0003/the-latin-language-evolved-over-time-masked-models/

I would rather define myself as a #GenAI sceptic (I've documented the reasons in my publications at https://www.zotero.org/mapto/publications), but here's what a professional adopter, i.e a founder of a GenAI startup (that I value lots) has to say about the state of technology. No need to say we diverge on the outlooks, but that's not the point here

"In a broader sense, however, today’s ruling is of a piece with this Court’s recent tendencies. “[R]ight when the Judiciary should be hunkering down to do all it can to preserve the law’s constraints,” the Court opts instead to make vindicating the rule of law and preventing manifestly injurious Government action as difficult as possible….. This is Calvinball jurisprudence with a twist. Calvinball has only one rule: There are no fixed rules. We seem to have two: that one, and this Administration always wins."

https://www.techdirt.com/2025/08/22/justice-jackson-correctly-defines-the-john-roberts-supreme-court-as-the-calvinball-court/

History keeps repeating, dictators keep pulling each-other's strings.

"In interviews for a book about his Middle East peace efforts, Trump, according to its author, used an expletive to describe the embattled prime minister — “Fuck him,” he reportedly said — and accused Netanyahu of disloyalty."

https://www.timesofisrael.com/trump-posts-clip-of-prof-calling-netanyahu-obsessive-about-getting-us-to-fight-iran/
#Trump #Netanyahu #JeffreySachs #Iran #Israel #Palestine #Lebanon #Syria #Iraq

Which part of this does not exactly hold for Katz and his involvement in bombing hospitals in Gaza?

Certainly that in the civilised world due process is part of life.

Context: https://www.theguardian.com/world/live/2025/jun/19/israel-iran-conflict-live-hospital-in-southern-israel-hit-in-iran-missile-strike-say-israeli-officials?page=with%3Ablock-6853e39e8f08a4f72f5f14a1#block-6853e39e8f08a4f72f5f14a1

This is how ruthless the Israeli military complex is

The sign talks of"protecting the state of Israel", remaining silent to the fact that this offensive "protection" actually costs hundreds of thousands of lives in Israel and the region.

Context:
https://www.theguardian.com/science/2025/jun/16/israeli-stands-paris-airshow-shut-down-french-government
#israel #palestina #lebanon #yemen #iran

It doesn't take much to create a #fedi alternative to #Linkedin. We just need structured profiles (featuring experience and education) and posts (that let people know if they could still apply or position is taken).