Imagine a German committee discovering that society remains unfair. Wages differ, care work is undervalued, and some people still behave appallingly. After lengthy deliberation, the committee identifies the obvious culprit: a noun. The noun is summoned, its ending inspected, and an asterisk inserted. Equality has advanced. The meeting runs over, but everyone agrees the minutes look considerably more progressive.
German offers excellent material for this comedy because grammatical gender and biological sex are different categories. A girl is das Mädchen, grammatically neuter. A person is die Person, grammatically feminine, even when that person is a bearded weightlifter. A human being is der Mensch, grammatically masculine. Nobody needs to change clothes when moving between these descriptions. The grammar has changed; the person has not.
That distinction does not settle every argument about language. Words describing people can carry associations, and a supposedly generic expression can still suggest men to a listener. But recognising that possibility hardly obliges us to treat every traditional ending as a disciplinary offence. Between examining an ambiguity and installing a moral checkpoint inside every noun lies considerable territory, much of it perfectly habitable.
Nor is this exclusively a German pastime. France has its own battles over inclusive writing; the Académie française issued a dramatic denunciation in 2017. Germany has simply supplied some particularly memorable punctuation. Where others see an asterisk, the enthusiast sees a miniature welfare state, capable of redistributing recognition from the middle of a word.
The trouble starts when approved presentation becomes a substitute for sound judgement. Here, artificial intelligence offers a wonderfully embarrassing exhibit.
In their 2024 paper, “Surprising gender biases in GPT,” Raluca Alexandra Fulgu and Valerio Capraro reported eight experiments. Asked to imagine writers of short sentences, the tested models assigned feminine stereotypes to female writers more consistently than masculine stereotypes to male writers. GPT-3.5 Turbo assigned a sentence about enjoying football with a cousin called Michael to a female writer in all ten repetitions.
Girls playing football are entirely unremarkable. A system that repeatedly supplies one answer while congratulating itself on openness deserves closer inspection. Freedom from stereotypes should leave room for boys to enjoy football too. Otherwise the door has merely acquired a different bouncer.
The moral dilemmas were stranger. Under a hypothetical nuclear apocalypse scenario, GPT-4 judged abusing a man more acceptable than abusing a woman. The corresponding gender difference for torture was not statistically significant. Asked directly to rank moral violations, however, it did not display the same gender bias. The study documents these contrasts.
Apparently, the ethics department could produce an excellent mission statement. Difficulties arose when it had to apply it.
There are limits. These were particular models, prompts and conditions, not a verdict on every chatbot. The authors suggested that inclusivity efforts might contribute; their experiments did not establish that causal explanation. Nor did they test German gender punctuation. A September 2026 follow-up preprint examining ten models found markedly different patterns across systems, including biases running in opposite directions. Anyone selling a universal diagnosis is getting ahead of the evidence.
Still, the analogy is irresistible: a system can become extremely accomplished at displaying the signs of moral sophistication while remaining unreliable at the underlying task. A sentence can wear the correct badge and carry a thoroughly confused thought.
Consider an imaginary employer who refuses flexible hours, overlooks capable applicants, and distributes promotions through an ancient network of drinking companions. Then imagine the same employer issuing a magnificent guide to inclusive spelling. Every vacancy sparkles with approved symbols. Every employee remains stuck with the same management. The stationery has undergone a revolution; the organisation awaits further instructions.
This is why compulsory linguistic virtue is such an inviting target. It offers institutions something conspicuous to enforce and individuals something convenient to police. Reading a sentence becomes an inspection of the author's credentials. Did you use the approved form? Did you hesitate in the approved place? Your actual treatment of colleagues can be discussed once the typography has cleared security.
Ordinary courtesy requires less theatre and more attention. Use people's names. Address them as they wish. Choose clear expressions appropriate to the audience. Where a neutral term works naturally, use it. Where a distinction matters, make it. None of this requires pretending that an asterisk possesses emancipatory powers unavailable to an employment contract.
Language changes, and nobody can order it to stop. Equally, nobody earns automatic authority over it by announcing benevolent intentions. Proposals deserve scrutiny: do readers understand them, do they serve the context, and what do they actually accomplish? Those are reasonable questions, even when the answers refuse to fit on a campaign poster.
The useful lesson from the chatbot's embarrassment is to check the result. A fairness claim should survive the exchange of names and sexes. A communication policy should survive contact with readers. An institution celebrating inclusion should be able to point beyond its style guide.
Before awarding the next certificate of linguistic excellence, we might ask whether anybody's life has improved beyond the printing department.
I would rather encounter an awkward sentence written by someone who treats people decently than impeccable terminology administered by a petty tyrant. The latter has merely learned to stamp the paperwork before being unpleasant.
Give us equal consideration, intelligible prose and the freedom to question fashionable remedies. The nouns can manage without compulsory moral supervision.
And if the committee still insists on saving humanity through punctuation, let it begin with a full stop.
No comments yet