Advanced English Skills

Language Log Good news for Tangutologists Tangutologists and Old Chinese aficionados take note: the whole 2024 Festschrift for M.V. Sofronov, produced on the occasion of his 90th birthday in 2019, is now available via C. Harbsmeier's academia page: https://t.co/hGUhgrgauE pic.twitter.com/zH7zxY0YjG — Wolfgang Behr (@behrwolf) June 27, 2024 Click on the illustration to embiggen, then follow the arrows to scroll horizontally through the pages of the table of contents. Many thanks to Wolfgang Behr for calling this publication to our attention via X and to Christoph Harbsmeier for making it readily available on academia.edu. I personally am particularly delighted to learn of this rich tribute to M. V. Sofronov because it constitutes an affirmation and validation for the hard work of my student, Nikita Kuzmin, who recently earned his Ph.D. in Tangut Studies and who has begun teaching it. Selected reading * "Polyglot Manchu emperor" (4/6/23) — especially in the comments * "Tangut beer" (10/13/18) * "Tangut workshop at Yale" (2/2/18) * M. V. Sofronov, "Chinese Philology and the Scripts of Central Asia", Sino-Platonic Papers, 30 (October, 1991), 1-10. [h.t. Geoff Wade] ➖ @EngSkills ➖

Language Log Stochastic parrots extended Philip Resnick, "Large Language Models are Biased Because They Are Large Language Models", arXiv.org 6/19/2024: This paper's primary goal is to provoke thoughtful discussion about the relationship between bias and fundamental properties of large language models. We do this by seeking to convince the reader that harmful biases are an inevitable consequence arising from the design of any large language model as LLMs are currently formulated. To the extent that this is true, it suggests that the problem of harmful bias cannot be properly addressed without a serious reconsideration of AI driven by LLMs, going back to the foundational assumptions underlying their design. Phil offer this example: A simple example demonstrates that this is not just an inconsequential observation about LLMs, but rather a fundamental property inherent in their design. Consider the word nurse, in its typical sense in English. Here are three statements that are statistically true about the concept that nurse denotes at this point in time and history. * * A nurse is a kind of healthcare worker. * A nurse is likely to wear blue clothing at work. * A nurse is likely to wear a dress to a formal occasion. The first of these is a fact about the meaning of the word and does not vary with context. To assert that someone is a nurse and that they do not work in healthcare is a contradiction. And for people, or AI, to make use of the fact that nurses are healthcare workers is normatively fine. The second statement is contingently true: it is true at the present time, but nothing about nurses makes it necessary. The statement is also normatively acceptable; for example, a person or an AI system classifying someone as nurse versus nonnurse is not engaging in harmful bias if it pays attention to the color of someone’s work clothes. The third statement is also contingently true in the same sense. However, it would be normatively unacceptable, in many contexts, to use that statistical fact in making inferences or decisions. For example, in speaking with well-dressed people at a party, it would be considered inappropriate to simply assume that a woman in a dress was more likely to be a nurse than a man in a suit, even if the assumption is statistically justifiable. Crucially, LLMs, as they are currently constituted and trained, have no basis for distinguishing among these three distinct statements about nurses. The representation of nurse in an LLM’s embedding space, and the contribution of nurse to contextual representations and inferences, makes no distinction between definitions versus contingent facts, nor between normatively acceptable versus unacceptable representations and inferences. It is distributionally observable, at the present time, that in large training samples the word nurse occurs far more frequently in the context of hospital than of theater, an observation grounded in its meaning. It is just as observable that the word nurse occurs far more frequently in sentences where the pronouns are she or her, but this observation is grounded only by contingencies in today’s society — a society that retains gender biases about the presumed role of women, about which kinds of jobs pay well or poorly, etc. (Cookson et al., 2023). LLMs create their representations entirely on the basis of observed distributions in language (Lenci, 2018), and they have no basis for distinguishing among these distributional observables. This extends idea implicit in Emily Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell, "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pp. 610-623. 2021. Some relevant past posts: "Stochastic parrots", 6/10/2021 "Copyrightsafe AI training", 7/23/2023 "Annals of AI bias", 9/23/2023 ➖ @EngSkills ➖

Funny Or Die (Youtube) A look back on alternative facts, a cabinet made up of idiots, and more in this #BestPresidencyEver At tonight's debate Trump is going to tell you how great things were when he was President. Check out our series #BestPresidencyEver for a reminder of what it was actually like. Subscribe now: https://www.youtube.com/c/funnyordie?sub_confirmation=1 Get more Funny Or Die ------------------------------- Facebook: https://www.facebook.com/funnyordie Twitter: https://twitter.com/funnyordie Instagram: http://instagram.com/funnyordie TikTok: https://www.tiktok.com/@funnyordie ➖ @EngSkills ➖

Funny Or Die (Youtube) Dad boner #shorts Subscribe now: https://www.youtube.com/c/funnyordie?sub_confirmation=1 Get more Funny Or Die ------------------------------- Facebook: https://www.facebook.com/funnyordie Twitter: https://twitter.com/funnyordie Instagram: http://instagram.com/funnyordie TikTok: https://www.tiktok.com/@funnyordie ➖ @EngSkills ➖

Learn English Through Football Euro 24 Football Phrase Day 14: Rank Outsider Euro 24 Football Language Phrase (Day 14): Day 14, and in this football language post for Euro 24 we look at the phrase, ‘rank outsider‘ as the last games of the group stage were played. Don’t forget we have hundreds more explanations of football language in our football glossary and we also have a page […] The post Euro 24 Football Phrase Day 14: Rank Outsider appeared first on Learn English Through Football. ➖ @EngSkills ➖

Language Log Pronunciation guides fail spectacularly ICYMI: Jonathan Edwards, "Mispronunciations spoil graduation at Ta-mul-may Jefferson University", WaPo 5/13/2024: The school is apologizing after viral footage showed a graduating Sarah announced as “Sigh-eer” while a Molly Elizabeth was pronounced “Mah-lee-nuh Zo-beth.” Sarah Virginia Brennan had never heard her first name mispronounced before. Then she walked across the commencement stage after graduating from Thomas Jefferson University. “Sigh-eer Oo-voon-jean-june Bree-nun,” a university official announced to friends and family of the graduates who had earned Bachelor of Science degrees from the university’s school of nursing. Brennan hesitated at first, unsure whether it was her turn to walk or someone else’s, she told The Washington Post. “I didn’t process how poorly she could do mine,” Brennan said. Then came Molly Elizabeth Camp. “Mah-lee-nuh Zo-beth Cahmp,” the official said. And finally there was Thomas Michael Canevari Jr. “Ta-mul-may,” the official started before Canevari cut her off. “Thomas,” he corrected her. It was the end of a painful six minutes at the commencement ceremony at the Philadelphia university on Thursday. Clips of the ceremony went viral over the weekend, racking up more than 20 million views and comparisons to Key and Peele’s 2012 skit “Substitute Teacher.” Anyone who's taught recently in U.S. post-secondary education will understand the reasons for adding pronunciation fields of some kind to the conventional spelling of student names. I could give dozens of examples from recent classes, where I had to ask how the name should be pronounced, and would have guessed wrong without asking. But it's clear that Jefferson University managed to screw up seriously in this case, though it's not clear to me exactly how. Apparently there were no regular spellings, and either the pronunciations were incompetently rendered, or the reader had no training in interpreting them, or both. The Tonight Show segment: Update — here's an image of one of the student name cards, showing that the problem was mainly announcer incompetence, and partly the failure of U.S. culture to create and teach a standard phonologically-transparent orthography: http://languagelog.ldc.upenn.edu/myl/JeffersonPronunciationCard.avif ➖ @EngSkills ➖

Word of the Day Word of the Day: expressly This word has appeared in 112 articles on NYTimes.com in the past year. Can you use it in a sentence? ➖ @EngSkills ➖

Idiom of the Day jet-setter A wealthy individual who travels globally, especially by private jet, to frequent fashionable resorts, social events, and the like. Watch the video ➖ @EngSkills ➖

cardinal

Word of the Day motley Definition: (adjective) Consisting of a haphazard assortment of different kinds. Synonyms: assorted, miscellaneous, mixed, sundry. Usage: The other occupants of the room, five in number, were all females, and they were still sleeping, piled high with a motley array of silks and furs. Discuss ➖ @EngSkills ➖