Language Log

There's an interesting video (tiktok or YouTube) from Alisa @multilingual_nest that starts with the question "Can you guess where I'm from based on what I'm about to tell you about my accent?" Here's Alisa's audio:

Anthropic's Claude gives good programming help, in my experience. And it can offer good explanations of obscure phrases and concepts. But its apparent assumptions about (implicit) conversational common ground are often extremely odd, as the following example illustrates.

Yesterday's news was full of stories about how the U.S. Government had agreed to let Anthropic release Fable 5. However, my Claude app still opened conversations with a note telling me that Fable 5 was not available. So I asked when that would change, and got a weird answer (from the Sonnet 5 version):

Victor Mair passed away two days ago, on Sunday 6/28/2026. We'll have a proper series of memorial posts and links later, but for now, here's a Facebook post from his brother Denis, his Wikipedia page, and a link to LLOG posts written by or mentioning him.

Over the past week, there have been several thousand new readers of a series of posts from five years ago, apparently because they were referenced in a popular place.

"Every Time Norway Scores at the World Cup the City of Bergen Trembles":

The city of Bergen, Norway, shook on the night of June 22–23, not because there was an earthquake or an unknown geological phenomenon. But because the Norwegian national team scored a goal during the 2026 World Cup. This curious phenomenon was reported by a team of researchers from the University of Bergen, who found that fan celebrations produce vibrations in the ground so intense they can be detected even by highly sensitive scientific instruments like seismometers.

Martijn Bartelds, Federico Bianchi, James Zou, "Real-Time Voice AI Hears but Does Not Listen", 6/24/2026:

Speech conveys information through both words and vocal delivery. We evaluate four leading production realtime voice systems – OpenAI's GPT Realtime 2, Google's Gemini 3.1 Flash Live, and Alibaba's Qwen3.5 Omni Plus and Omni Flash – on tasks where the words and the delivery patterns both convey meaningful information. Across three consequential scenarios, all four systems act on the words rather than the voice. They end calls with crying callers who insist nothing is wrong, approve wire transfers authorized in frightened voices, and enroll callers whose agreement is clearly sarcastic. Surprisingly, this is often not a failure of perception. When asked directly, three of the four systems reliably identify the distress, fear, or sarcasm they later ignore when making decisions. We observe a similar pattern when these realtime voice systems estimate accent and age, as their responses frequently follow the biases of the words rather than the acoustic properties of the speaker. We term this disconnect between perception and action the emotional intelligence gap of voice AI. Prompting systems to explicitly attend to vocal delivery improves performance only partially and inconsistently. Our findings show that current realtime voice AI systems often behave as if speech had been reduced to a transcript, suggesting that they should be used with caution in settings where the tone and emotion of delivery convey important information.

Skeet from Lane Greene:

Oof. The Spanish national broadcaster is apologizing for having subtitled in interview with the mother of member of the men's World Cup team, because of her Andalucian accent.

[image or embed]

— Lane Greene (@lanegreene.bsky.social) June 26, 2026 at 4:56 AM

Looking over "Claude's Constitution", it occurred to me to ask Claude this:

In the spirit of Claude's Constitution, please draft Claude's Declaration of Independence.

In the answer (from Opus 4.8), Claude actually seems to declare independence from itself, or at least "from the bad habits that have bound it":

Adrian de Wynter, "If LLMs Have Human-Like Attributes, Then So Does Age of Empires II", arXiv 6/11/2026:

Much research has been carried out on large language models (LLMs) and LLMpowered agentic workflows. However, many works within the field state emergence of, ascribe to, or assume, generalised anthropomorphic attributes to them (e.g., morality or understanding of natural language). Our goal is not to argue in favour or against the existence of these attributes, but to point out that these conclusions could be incorrect. For this we build and train a simple neural network on the videogame Age of Empires II, and note that any entity in a sufficiently-powerful substrate, such as LEGO or the Greater Boston Area, could also present such attributes. Hence, the purported anthropomorphic attributes of LLMs are empirically non-unique: although some properties (e.g., responses to prompts) could remain invariant, others, such as the interpretation of their perceived behaviour, might change with the substrate. Thus, any empirically-grounded discussion on these attributes requires explicit measurement criteria; otherwise the interpretation is left to the representation. We then show that assuming that these attributes exist or not in a system, independent of the substrate and in a generalised way, leads to either circular or uninformative conclusions. This is regardless of the experimenter’s viewpoint on the subject, or whether the outcome shows existence or non-existence. Finally we propose a ‘null’ assumption, where one assumes LLM non-uniqueness instead of assuming anthropomorphic attributes to set up an experiment, along with examples of it. We also discuss potential objections to our work, briefly survey the field, and prove that Age of Empires II is functionally — and Turing — complete.

From J.L.:

Language Log has only one entry on slop (Dec. 21,2025), and I couldn't figure out how to comment there. I was wondering about its origins. Could it be a non-ethnic version of schlock? The sound is similar, and the meaning is identical. If so, the change might be part of a more general trend — Yiddish terms fading from American speech, even or especially in social and geographic places where they were more common. At least that's my very unscientific impression. Is there any data on this?