Updated 13 October 2026. The retrospective archive date and later literature are documented in the changelog.

Consider a sentence on a school website: Die Lehrer treffen sich am Freitag — “The teachers meet on Friday.”

The writer may mean the whole teaching staff. The staff may include people of several genders. Nothing in the sentence explicitly says that only men attend.

Now add: Einige der Frauen bringen Unterrichtsmaterial mit — “Some of the women bring teaching materials.” Does that continuation feel as straightforward as the same sentence about Männer, men?

These are examples I constructed, not items reproduced from an experiment. They illustrate a question that can be tested. Knowing whom a writer intends to include and processing those people equally readily are different things.

The disagreement about the generisches Maskulinum — the generic masculine — becomes more interesting once those two questions are kept apart.

A Grammatical Category Is Not a Person

German grammatical gender classifies nouns and governs agreement. It is not a biological description of everything named by them. Die Person is grammatically feminine and can refer to a man; das Mitglied, “the member”, is neuter and can refer to a woman. With many occupational nouns, however, masculine and feminine forms also distinguish reference to people: Lehrer and Lehrerin, for example. The grammatical and referential systems overlap without being identical [[1]].

There is a further complication in the plural. Die is the article for plural nouns across grammatical genders. The die in die Lehrer does not turn the role noun into a feminine form. The relevant contrast is between Lehrer and Lehrerinnen, together with their conventions of use.

A masculine role noun can name men specifically or be used with an inclusive intention. Moreover, “generic” in this debate can mean gender-inclusive reference to a particular mixed group. It need not mean a generalisation about an entire class of people. Our Friday meeting concerns a specific staff, while Lehrer benötigen Geduld — “Teachers need patience” — is a general statement. The ambiguity of gender reference is a separate dimension.

An analysis of grammatical meaning can establish that inclusive use is possible. An experiment asks what participants do with a form in a specified context. Neither question substitutes for the other.

What the Famous Experiment Asked

In 2008 Gygax and colleagues compared English, French, and German, recruiting 36 university students per language. Two participants were excluded. Readers judged whether a sentence mentioning women or men sensibly continued a preceding sentence about a group. The role nouns covered male-associated, female-associated, and relatively balanced occupations and roles; incompatible filler pairs required rejection [[2]].

The measurements were the proportion of accepted continuations and response times for accepted continuations. French and German masculine forms favoured continuations about men; occupational stereotypes did not account for that pattern in those samples. English responses were sensitive to stereotypes. This was evidence about integration of subsequent gender information, not a direct count of figures in a reader’s private mental picture [[2]].

The often repeated phrase “women are invisible” compresses that result too much. The experiment did not show that participants were incapable of understanding a continuation about women. It measured an asymmetry in accepting and processing it.

An awkward continuation can reflect several things: an initial expectation, a particular reading of an ambiguous noun, pragmatic assumptions about the group, or features of the decision task. Calling the result a male bias describes the direction of the measured difference. It does not identify every cognitive operation responsible for it.

The Effect Survives a Different Task

A politically useful result deserves replication as much as an inconvenient one. Small student samples from 2008 cannot by themselves settle contemporary German usage.

Brohmer, Hofer, and colleagues’ registered 2024 replication of Stahlberg, Sczesny, and Braun’s 2001 celebrity-naming study involved twelve laboratories. Participants named up to three celebrities in each of six categories, with randomly assigned masculine, neutralised, paired, internal-capital-I, or asterisk instructions. The extended analyses used 2,697 participants; the narrower close replication used 1,479. Inclusive alternatives elicited more women than masculine and neutralised wording. Adjustment for perceived representation of women in each category did not remove the main contrast [[3]].

In the close replication’s four original categories, with twelve possible names, estimated mean female-name totals were 2.47 for masculine wording, 3.60 for pairs, and 5.52 for capital-I. The pair-versus-masculine standardised difference was 0.51, with a 95% confidence interval of 0.38–0.63. Those estimates concern the specified retrieval task [[3]].

The frequently cited 0.84 pools the paired and capital-I conditions against masculine wording in the close replication: the across-laboratory meta-analysis gives a 90% confidence interval of 0.67–1.01. It is neither the pair-only comparison nor an estimate for the asterisk in all 2,697 participants [[3]].

The neutralised condition matters: explicit inclusion also differed from wording without gender information. Alternatives can actively cue women, rather than merely remove a masculine cue [[3]].

This is a different dependent variable from continuation acceptance: which names become available and are selected for an answer. Random assignment makes a causal interpretation of the wording difference more credible within this task. It does not convert celebrity retrieval into a measurement of hiring discrimination.

The studies converge on a bounded claim: wording can affect the relative accessibility of women and men in responses. They leave the magnitude in other situations to be measured.

Intention Does Not Specify the Result

The strongest defence of the generic masculine is straightforward: competent speakers know that it can include women. That defence establishes a convention and a communicative intention. The empirical issue is whether knowing the convention eliminates the processing asymmetry.

Körner, Abraham, Rummer, and Strack tested German sentence continuations in 2022. Participants in the masculine and asterisk conditions were explicitly informed that those forms included women and men. Nevertheless, masculine wording favoured male continuations. Pair forms produced comparatively balanced responses. The asterisk produced a female bias in acceptance in both experiments, with a response-time bias detected only in the second [[4]].

That pattern challenges both a simple “they just misunderstood the instructions” explanation and a simple “the asterisk makes representations neutral” explanation. It also illustrates why acceptance and latency should be reported separately: they are related measurements, not interchangeable readings from one bias meter.

Frequency is another reasonable objection. If men are more common in a profession or more prominent in media coverage, retrieving more male examples may reflect experience. Comparisons within the same occupational category help here: the question is whether changing the wording changes the responses while the category remains fixed. A female-majority occupation need not erase the association of a masculine form; conversely, a neutral noun need not erase a familiar occupational stereotype.

No account requires women and men to have different kinds of minds. Participants have histories of language use, familiarity, and social experience. A group average is not a diagnosis of every reader, and a participant’s recorded gender is not a complete account of those histories.

There Is More Than One Alternative

Lehrerinnen und Lehrer, Lehrkräfte, and Lehrer*innen are different interventions. The first explicitly names two forms. The second uses an occupational term without that masculine–feminine distinction in reference, although the noun still has grammatical gender. The third changes the written morphology and marks a convention intended to include people beyond the woman–man distinction.

They can also change emphasis and register. A replacement must fit the intended meaning, case, number, and sentence. Treating every inclusive form as one experimental condition conceals those choices.

A 2025 electrophysiological study by Glim, Körner, and Rummer adds a complication. In another sentence-pair task, masculine wording was followed by a larger P600 response to references to women; after asterisk wording the larger response occurred for references to men, in a somewhat different spatial and temporal pattern. The authors interpreted this as evidence of unequal processing demands for both forms [[5]].

A scalp-recorded electrical response is no ballot on preferred spelling. The finding is relevant because it weakens an unrestricted neutrality claim for the asterisk. It does not show that a female processing bias is socially identical to a male one, or that either form excludes a person from a job.

There is also a measurement gap. Testing only continuations about women and men cannot establish how readily nonbinary people are represented. A symbol’s intended scope, the familiarity of its convention, and a reader’s actual interpretation need separate evidence. Equal performance on two binary categories would still leave that third question unanswered.

Readability Is Several Measurements

Reading time, word recognition, subjective ease, comprehension-test performance, and liking a text are different outcomes. “Harder to read” can refer to any of them.

Friedrich and colleagues’ 2021 experiments illustrate the danger of combining them. A board-game text containing predominantly plural asterisk forms produced no impairment on their measures and some favourable ratings. A contact-sport text with many more complex singular constructions received lower comprehensibility ratings. The texts differed in several ways, so those experiments did not isolate number as the cause. Their comprehensibility measures were subjective ratings; they were not a direct test of how much factual content readers retained [[6]].

A 2024 follow-up compared masculine, plural-asterisk, and singular-asterisk versions in two experiments. Singular forms lowered comprehensibility ratings in both; plural forms did so only in the first. That supplies more controlled evidence that the construction matters, while retaining the distinction between rated ease and tested retention [[10]].

Zacharski, Kruppa, and Ferstl’s 2025 study used lexical decisions: participants decided whether a displayed string was a German word. Among those accepting asterisk nouns as words, students processed them comparably to masculine and feminine forms. Non-students were initially slower and accepted them less often; their response-time disadvantage diminished during the experiment. That supports a role for familiarity in this task. It does not demonstrate equivalent comprehension of a complete text, or establish accessibility for every reader [[7]].

For teaching material, I would want both comprehension and usability tested with the people who actually use it. A familiar spelling can be effortless to identify while supplying a skewed cue about the people discussed. An unfamiliar spelling can convey an inclusive intention while interrupting reading. Good editing has to consider both.

This is a recommendation about evaluation, not a finding that every classroom should adopt one orthographic device.

Familiarity Can Change Some Effects

It would be odd to treat the effects of conventions learned through language as permanently fixed. But “people will get used to it” is also an empirical claim.

The literature update of 13 October 2026 adds a study by Körner, Hofmann, Glim, Härtl, and Rummer, published that June, with 272 participants. An initial task made the asterisk’s inclusive use salient by linking it equally to women and men. The manipulation did not change subsequent celebrity retrieval or continuation error rates. It did remove the female response-speed bias in the subsequent sentence task. One of three measures supported the predicted change [[8]].

The result provides limited evidence for malleability. It does not establish that sustained exposure removes every bias, nor that the same procedure removes the masculine form’s male bias: that was not the intervention tested.

This is also a useful response to the objection that experiments create artificial conditions. They do. Here, changing part of the preceding context changed one later outcome. Context is part of what needs explaining. A scientific account should specify how cues work together, rather than require a noun to generate the same response in every sentence.

From a Continuation to an Opportunity

Why care about a fraction of a second or a celebrity list?

Because institutions routinely ask people to imagine candidates, experts, students, and colleagues. A wording effect on who comes to mind supplies a plausible mechanism for uneven attention. Whether it changes an actual decision depends on what happens next: available applicants, explicit criteria, deliberation, and opportunities to revise an initial impression.

Evidence closer to education exists. Vervecken and colleagues tested occupational descriptions with 222 French-speaking adolescents in Switzerland, excluding one from the analyses. Pair forms reduced differences in attributed success between women and men in gender-associated occupations and affected warmth judgments. They found no effect of language form on competence judgments. The outcomes were immediate perceptions, not subsequent occupational choices [[9]].

That null finding belongs in the account. A linguistic intervention can affect one judgment without shifting every stereotype. French adolescent ratings also cannot provide a numerical estimate for the effect of German asterisks on university enrolment.

The progression from processing to opportunity is a research programme, not a conversion factor. No number of milliseconds can be multiplied by a universal constant to obtain lost careers.

What I Would Change in a School Text

My judgment is that a school should take seriously whether its descriptions make all the people it intends to address readily imaginable. That is a normative commitment. The experiments help assess ways of meeting it; they do not choose the commitment for us.

For the opening sentence, Die Lehrkräfte treffen sich am Freitag is a natural alternative. It preserves the reference to teachers and avoids asking one role noun to do both specifically male and inclusive work. Elsewhere a pair form may make women’s participation explicit, while an asterisk may communicate a broader inclusive convention. The best choice depends on purpose and readers; it cannot be read from one study’s significance test.

There is a computational follow-up, but it is a separate article. Measuring how a particular tokeniser splits Lehrer*innen would establish its segmentation. As the strawberry article explains, a representation boundary does not identify the model’s hidden reasoning. Extra tokens alone would not show gender bias, poor translation, or exclusion. Those require controlled output tests of a specified system.

The useful feminist question here is precise enough to investigate: when an institution says everyone is included, do its habitual descriptions help readers understand that inclusion?

The meeting is on Friday. The wording should help people find themselves in the room.

Literature checked through 2026-10-13. Study designs and reported outcomes were checked against primary publications; no participant-level data were reanalysed and no LLM experiment was conducted.

References

[1] Sczesny, S., Formanowicz, M., & Moser, F. (2016). Can gender-fair language reduce gender stereotyping and discrimination? Frontiers in Psychology, 7, 25. https://doi.org/10.3389/fpsyg.2016.00025.

[2] Gygax, P., Gabriel, U., Sarrasin, O., Oakhill, J., & Garnham, A. (2008). Generically intended, but specifically interpreted: When beauticians, musicians, and mechanics are all men. Language and Cognitive Processes, 23(3), 464–485. https://doi.org/10.1080/01690960701702035. Institutional manuscript record.

[3] Brohmer, H., Hofer, G., et al. (2024). Effects of the generic masculine and its alternatives in Germanophone countries: A multi-lab replication and extension of Stahlberg, Sczesny, and Braun (2001). International Review of Social Psychology, 37(1), 17, 1–25. https://doi.org/10.5334/irsp.522. Publisher PDF.

[4] Körner, A., Abraham, B., Rummer, R., & Strack, F. (2022). Gender representations elicited by the gender star form. Journal of Language and Social Psychology, 41(5), 553–571. https://doi.org/10.1177/0261927X221080181.

[5] Glim, S., Körner, A., & Rummer, R. (2025). An ERP-based comparison of gender representations elicited by generic masculine role nouns and the German gender star form. Journal of Neurolinguistics, 73, 101231. https://doi.org/10.1016/j.jneuroling.2024.101231.

[6] Friedrich, M. C. G., Drößler, V., Oberlehberg, N., & Heise, E. (2021). The influence of the gender asterisk (“Gendersternchen”) on comprehensibility and interest. Frontiers in Psychology, 12, 760062. https://doi.org/10.3389/fpsyg.2021.760062.

[7] Zacharski, L., Kruppa, A., & Ferstl, E. C. (2025). The readability of the non-binary gender star in German: Evidence from a lexical decision task. Social Psychological Bulletin, 20, e13719. https://doi.org/10.32872/spb.13719.

[8] Körner, A., Hofmann, P., Glim, S., Härtl, H., & Rummer, R. (2026). Do experiences influence gender representations elicited by the gender star form? Applied Psycholinguistics, 47, e27. https://doi.org/10.1017/S0142716426100654.

[9] Vervecken, D., Gygax, P. M., Gabriel, U., Guillod, M., & Hannover, B. (2015). Warm-hearted businessmen, competitive housewives? Effects of gender-fair language on adolescents’ perceptions of occupations. Frontiers in Psychology, 6, 1437. https://doi.org/10.3389/fpsyg.2015.01437.

[10] Friedrich, M. C. G., Gajewski, S., Hagenberg, K., Wenz, C., & Heise, E. (2024). Does the gender asterisk (“Gendersternchen”) as a special form of gender-fair language impair comprehensibility? Discourse Processes, 61(9), 439–461. https://doi.org/10.1080/0163853X.2024.2362027.

Changelog

  • 2026-10-13 — Retrospective Archive and Later Literature: Assigned 15 September 2025 as the archive date, beside the blog’s education and language discussions. This version was prepared and checked in October 2026; the archive date does not claim that it was online in September 2025. The June 2026 study by Körner and colleagues is a later addition: it changes one response-speed result while leaving retrieval and error rates unchanged. The account therefore supports limited malleability, not a claim that familiarity removes every representational bias. This review also checked the 2025 ERP and lexical-decision papers and separated their different outcomes, and clarified the distinct effect-size comparisons in the 2024 replication. Those refinements preserve the distinction between laboratory responses and real-world opportunities.
  • 2026-10-13 — Effect-Size Review: Identified the pooled contrast behind the widely reported 0.84 effect size and distinguished it from the paired-form comparison and the larger extended sample.
  • 2026-10-13 — Research Review: Distinguished intended reference from measured interpretation, compared separate inclusive-language interventions, and marked the limits of binary tasks and laboratory outcomes.