The "Small World of Words" German Free-Association Norms

Dirk U. Wulff; Kaidi L\~oo; Rui Mata; Samuel Aeschbach; Simon De Deyne

arxiv: 2604.19620 · v1 · submitted 2026-04-21 · 💻 cs.CL

The "Small World of Words" German Free-Association Norms

Samuel Aeschbach , Rui Mata , Kaidi L\~oo , Simon De Deyne , Dirk U. Wulff This is my paper

Pith reviewed 2026-05-10 03:13 UTC · model grok-4.3

classification 💻 cs.CL

keywords free association normsGerman languagepsycholinguisticssemantic networkslexical decisionword associationscross-linguistic comparison

0 comments

The pith

German free-association norms for 5,877 words predict performance in lexical decision tasks, relatedness judgments, and word ratings.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces the largest collection of free-association norms for German as part of the Small World of Words project. These norms are validated by demonstrating robust prediction of results from three standard psycholinguistic tasks involving word recognition and semantic relations. The work also shows the new data aligns well with prior German resources while highlighting both common and language-specific patterns when compared across languages. This supplies researchers with an empirical foundation for examining how German speakers mentally connect words.

Core claim

The authors collected free-association responses for 5,877 German cue words and demonstrated that these norms robustly predict performance in lexical decision tasks, relatedness judgments, and psycholinguistic word ratings. The SWOW-DE dataset compares favorably with existing German resources and shows both shared and language-specific association patterns in preliminary cross-linguistic comparisons.

What carries the argument

The SWOW-DE free-association norms, which record people's spontaneous word associations to German cues and function as data to model semantic structure and forecast behavioral outcomes in cognitive experiments.

If this is right

SWOW-DE enables more accurate modeling of semantic processing for German speakers.
The resource supports cross-linguistic studies by allowing comparisons of association patterns across languages.
Researchers can use the norms to validate or extend other German psycholinguistic datasets.
Future work on cultural influences in word associations can build directly on this collection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The norms might help improve German-language AI systems by adding association-based semantic knowledge.
Applying the same validation approach to other languages could standardize cross-cultural semantic research.
Investigating how these associations change with age or dialect within German speakers would extend the findings.

Load-bearing premise

The online participant sample and chosen cues produce representative German associations, and the preprocessing steps remove noise without altering the underlying associative structure.

What would settle it

New free-association data gathered from an offline, demographically matched German sample that shows substantially weaker correlations with independent lexical decision latencies or relatedness ratings would falsify the predictive robustness.

read the original abstract

Free-association norms provide essential empirical data for investigating linguistic, semantic, and cultural phenomena in the cognitive sciences. Although large-scale norms exist for languages such as English, Dutch, Spanish, and Mandarin Chinese, no comparable resource has been available for German. To address this gap, we present free-association norms for 5,877 German cue words as part of the German version of the multilingual Small World of Words (SWOW) project. We describe the data collection procedures, participant characteristics, and our comprehensive preprocessing pipeline before introducing the resulting SWOW-DE data set. Using data from three established psycholinguistic paradigms, we show that SWOW-DE norms robustly predict performance in lexical decision tasks, relatedness judgments, and psycholinguistic word ratings. Furthermore, we demonstrate that SWOW-DE responses compare favorably with existing German resources and provide a preliminary cross-linguistic comparison revealing both shared and language-specific association patterns, highlighting promising directions for future research. Overall, SWOW-DE represents the largest collection of German free associations to date and offers a unique resource for linguistic, psychological, and cross-cultural research.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The real contribution is the new large-scale German free-association dataset itself; the validation claims look preliminary and could be tightened against lexical confounds.

read the letter

The main thing to know is that this paper creates SWOW-DE, free-association norms for 5,877 German cue words, which is the first resource of this size for the language. They followed the established SWOW collection and preprocessing approach, then tested whether the resulting norms predict lexical decision performance, relatedness judgments, and word ratings, plus some comparisons to prior German data and other languages. That fills a practical gap for anyone who needs German-specific association structure rather than relying on English or Dutch proxies. The data release and the cross-linguistic notes are the parts that actually move the field forward. The validation section is where it gets thinner. The abstract calls the predictions robust, but without seeing explicit controls for frequency, length, or neighborhood density in the regressions, it's unclear how much unique variance the association measures explain versus those standard lexical variables. Lexical decision times in particular are heavily driven by frequency, so any claim of robustness needs incremental validity checks to hold up. The participant details and exclusion rules also matter for representativeness, and those are only sketched at the abstract level. Even so, the core resource still has value on its own. This is for psycholinguists and computational researchers who work with German or multilingual semantics and need ready association data. A reader who wants to plug the norms into models or run follow-up experiments will get something usable. It deserves a serious referee to check the collection pipeline, the statistical controls, and data accessibility rather than a desk reject.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces SWOW-DE, a large-scale free-association norm dataset for 5,877 German cue words collected via the multilingual Small World of Words project. It details participant recruitment, data collection procedures, and a preprocessing pipeline, then validates the norms by showing they predict performance in lexical decision tasks, relatedness judgments, and psycholinguistic word ratings from three established paradigms. The norms are shown to compare favorably with existing German resources, with a preliminary cross-linguistic comparison highlighting shared and language-specific patterns.

Significance. If the predictive validations hold after appropriate controls, SWOW-DE would fill a critical gap as the largest German free-association resource, supporting research on semantic networks, lexical processing, and cross-cultural psycholinguistics. The multi-paradigm validation and cross-linguistic angle add utility beyond simple norm collection.

major comments (2)

[Abstract and validation sections] Abstract and validation sections: the claim that SWOW-DE norms 'robustly predict' performance in lexical decision tasks, relatedness judgments, and word ratings does not address controls for lexical confounds. Lexical decision RTs are dominated by frequency, length, and neighborhood density; without reporting incremental validity (e.g., regressions adding association measures after these covariates) or partial correlations, the unique contribution of the free-association structure cannot be established and the 'robust' prediction claim is undermined.
[Methods and Results] Methods and Results: no sample sizes, exclusion criteria, statistical details, or error bars are provided for the participant sample or the three validation analyses. This absence prevents assessment of whether the online sample and cue selection yield representative norms or whether the reported predictions are statistically reliable.

minor comments (2)

[Abstract] Abstract: the phrase 'comprehensive preprocessing pipeline' is used without listing key steps (e.g., response cleaning, cue selection criteria); a one-sentence summary would improve readability.
[Cross-linguistic comparison] Cross-linguistic comparison: the preliminary analysis is mentioned but lacks quantitative metrics or example pairs of shared vs. language-specific associations; a small table would clarify the patterns.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback, which has helped us identify areas where the manuscript can be strengthened. We address each major comment below and describe the revisions we will implement.

read point-by-point responses

Referee: [Abstract and validation sections] Abstract and validation sections: the claim that SWOW-DE norms 'robustly predict' performance in lexical decision tasks, relatedness judgments, and word ratings does not address controls for lexical confounds. Lexical decision RTs are dominated by frequency, length, and neighborhood density; without reporting incremental validity (e.g., regressions adding association measures after these covariates) or partial correlations, the unique contribution of the free-association structure cannot be established and the 'robust' prediction claim is undermined.

Authors: We agree that the current validation analyses do not sufficiently control for lexical confounds such as word frequency, length, and neighborhood density, which limits the strength of our 'robustly predict' claim. In the revised manuscript, we will add multiple regression analyses for each validation paradigm that first enter the standard lexical covariates and then assess the incremental contribution of the SWOW-DE association measures (e.g., via change in R² and partial correlations). We will update the abstract, results, and discussion sections to report these new findings and qualify the prediction claims accordingly. This revision directly addresses the concern and will provide clearer evidence of the unique value of the free-association norms. revision: yes
Referee: [Methods and Results] Methods and Results: no sample sizes, exclusion criteria, statistical details, or error bars are provided for the participant sample or the three validation analyses. This absence prevents assessment of whether the online sample and cue selection yield representative norms or whether the reported predictions are statistically reliable.

Authors: We acknowledge that the omission of these details reduces transparency and makes it difficult to evaluate the reliability of the norms and validations. In the revised manuscript, we will expand the Methods section to report the full participant sample size, recruitment details, demographic characteristics, and all exclusion criteria applied. For each of the three validation analyses, we will add the relevant sample sizes (words and observations), exact statistical procedures, test statistics, p-values, effect sizes, and error bars on all figures. These additions will be presented in both text and tables to allow readers to assess statistical reliability and representativeness. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical data collection with external validation

full rationale

The paper collects new free-association norms from participants, applies preprocessing, and validates the resulting dataset by correlating it with performance measures from three independent psycholinguistic paradigms (lexical decision tasks, relatedness judgments, word ratings). These validation targets are drawn from separate established resources and are not derived from or fitted to the SWOW-DE data itself. No equations, parameter fitting, self-definitional loops, or load-bearing self-citations appear in the derivation chain. The central claim reduces to an empirical correlation between independently collected datasets, which is self-contained and non-circular by the stated criteria.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work rests on standard assumptions of psycholinguistic norm collection rather than new mathematical constructs or fitted parameters.

axioms (2)

domain assumption Online participants provide honest and representative free associations for German cue words.
Implicit in any crowdsourced norm collection; no explicit test of representativeness is described in the abstract.
domain assumption Standard preprocessing steps preserve the semantic signal without introducing systematic bias.
Preprocessing pipeline is mentioned but not detailed; validity is assumed rather than demonstrated.

pith-pipeline@v0.9.0 · 5504 in / 1207 out tokens · 32886 ms · 2026-05-10T03:13:44.160873+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

[1]

LLM World of Words

https://doi.org/10.1037/a0038693 Abramski, K., Improta, R., Rossetti, G., & Stella, M. (2025). The “LLM World of Words” English free associ- ation norms generated by large language models. Scientific Data,12(1), Article 803. https://doi.org/ 10.1038/s41597-025-05156-9 Aeschbach, S., Mata, R., & Wulff, D. U. (2025a). Mapping mental representations with fre...

work page doi:10.1037/a0038693 2025
[2]

https://doi.org/10.3758/s13428-018-1115-7 De Deyne, S., Perfors, A., & Navarro, D. J. (2017). Predict- ing human similarity judgments with distributional models: The value of word associations.Proceed- ings of the Twenty-Sixth International Joint Confer- ence on Artificial Intelligence, 4806–4810. https:// doi.org/10.24963/ijcai.2017/671 Dubossarsky, H., ...

work page doi:10.3758/s13428-018-1115-7 2017
[3]

(2018, March).Learning word vectors for 157 languages

https://doi.org/10.3758/BF03211553 16AESCHBACH Grave, E., Bojanowski, P., Gupta, P., Joulin, A., & Mikolov, T. (2018, March).Learning word vectors for 157 languages. arXiv. https://doi.org/10.48550/arXiv. 1802.06893 Gurevych, I. (2005). Using the structure of a concep- tual network in computing semantic relatedness. In R. Dale, K.-F. Wong, J. Su, & O. Y ....

work page doi:10.3758/bf03211553 2018
[4]

Computer Graphics Forum 31, 2pt2 (2012), 519–528

https://doi.org/10.1111/j.1467- 9280.2009. 02365.x hunspell. (2022).Hunspell. (Version 1.7.2) [Source code]. Github. https://github.com/hunspell/hunspell Hussain, Z., Binz, M., Mata, R., & Wulff, D. U. (2024). A tutorial on open-source large language models for behavioral science.Behavior Research Methods, 56(8), 8214–8237. https://doi.org/10.3758/s13428-...

work page doi:10.1111/j.1467- 2009
[5]

unknown word

https://doi.org/10.1037/0033-2909.87.2.245 Steyvers, M., & Tenenbaum, J. B. (2005). The large-scale structure of semantic networks: Statistical analy- ses and a model of semantic growth.Cognitive Sci- ence,29(1), 41–78. https :/ /doi . org/10 . 1207/ s15516709cog2901_3 The Stanford Natural Language Processing Group. (n.d.). Stanford log-linear part-of-spe...

work page doi:10.1037/0033-2909.87.2.245 2005
[6]

Beispiele: Cue: Mensch; Response: Homo sapiens; Korrigiert: Homo sapiens Cue: Theorie; Response: The Big Bang Theory; Korrigiert: The Big Bang Theory

Wenn eine Response a) richtig geschrieben ist auf Deutsch, b) richtig geschrieben ist in einer anderen Sprache und kein typisches, deutsches falsch geschriebenes Wort ist, oder c) ein korrekt geschriebener Eigenname ist: Die originale Response verwenden. Beispiele: Cue: Mensch; Response: Homo sapiens; Korrigiert: Homo sapiens Cue: Theorie; Response: The B...

work page
[7]

Beispiele: Cue: Argument; Response: Diskussoin; Korrigiert: Diskussion Cue: Marketing; Response: Werbungh; Korrigiert: Werbung

Wenn eine Response falsch geschrieben ist und (in Betracht des Cues und der anderen Assoziationen) die richtige Schreibweise zugerordnet werden kann: Die korrekt geschriebene Response verwenden. Beispiele: Cue: Argument; Response: Diskussoin; Korrigiert: Diskussion Cue: Marketing; Response: Werbungh; Korrigiert: Werbung

work page
[8]

Beispiele: Cue: hören; Response: sehen, fühlen, riechen; Korrigiert: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Korrigiert: feinkörnig

Wenn eine Response Wortkonstrukte enthält die auf mehrere Responses in einem Antwortfeld hinweisen: Die erste Response verwenden. Beispiele: Cue: hören; Response: sehen, fühlen, riechen; Korrigiert: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Korrigiert: feinkörnig

work page
[9]

Beispiele: Cue: lokal; Response: Lokal/Bar; Korrigiert: Bar

Wenn eine Assoziation Wortkonstrukte enthält, die den Cue wiederholen und eine zusätzliche, eigenständige Komponente enthalten: Die korrekt geschriebene eigenständige Komponente verwenden. Beispiele: Cue: lokal; Response: Lokal/Bar; Korrigiert: Bar

work page
[10]

Wenn eine Response ein unvollständiges Wort ist aber in Kombination mit dem Cue Sinn macht: Die sinnvolle Kombination verwenden. Beispiele: Cue: Gebiet; Response: Hohheits; Korrigiert: Hoheitsgebiet Cue: notwendig; Response: keit; Korrigiert: Notwendigkeit English translation of the system prompt: You help me correct potentially misspelled responses in a ...

work page
[11]

Examples: Cue: Mensch; Response: Homo sapiens; Corrected: Homo sapiens Cue: Theorie; Response: The Big Bang Theory; Corrected: The Big Bang Theory

If a response a) is correctly spelled in German, b) is correctly spelled in another language and is not a typical, misspelled German word, or c) is a correctly spelled proper name: Use the original response. Examples: Cue: Mensch; Response: Homo sapiens; Corrected: Homo sapiens Cue: Theorie; Response: The Big Bang Theory; Corrected: The Big Bang Theory

work page
[12]

SWOW-DE21 Examples: Cue: Argument; Response: Diskussoin; Corrected: Diskussion Cue: Marketing; Response: Werbungh; Corrected: Werbung

If a response is misspelled and (considering the cue and the other associations) the correct spelling can be assigned: Use the correctly spelled response. SWOW-DE21 Examples: Cue: Argument; Response: Diskussoin; Corrected: Diskussion Cue: Marketing; Response: Werbungh; Corrected: Werbung

work page
[13]

Examples: Cue: hören; Response: sehen, fühlen, riechen; Corrected: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Corrected: feinkörnig

If a response contains word constructs that indicate multiple responses in one answer field: Use the first response. Examples: Cue: hören; Response: sehen, fühlen, riechen; Corrected: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Corrected: feinkörnig

work page
[14]

Examples: Cue: lokal; Response: Lokal/Bar; Corrected: Bar

If an association contains word constructs that repeat the cue and contain an additional, independent component: Use the correctly spelled independent component. Examples: Cue: lokal; Response: Lokal/Bar; Corrected: Bar

work page
[15]

If a response is an incomplete word but makes sense in combination with the cue: Use the meaningful combination. Examples: Cue: Gebiet; Response: Hohheits; Corrected: Hoheitsgebiet Cue: notwendig; Response: keit; Corrected: Notwendigkeit User Prompt For each incorrectly spelled response, the following prompt template was used to obtain a correction. Respo...

work page 2020

[1] [1]

LLM World of Words

https://doi.org/10.1037/a0038693 Abramski, K., Improta, R., Rossetti, G., & Stella, M. (2025). The “LLM World of Words” English free associ- ation norms generated by large language models. Scientific Data,12(1), Article 803. https://doi.org/ 10.1038/s41597-025-05156-9 Aeschbach, S., Mata, R., & Wulff, D. U. (2025a). Mapping mental representations with fre...

work page doi:10.1037/a0038693 2025

[2] [2]

https://doi.org/10.3758/s13428-018-1115-7 De Deyne, S., Perfors, A., & Navarro, D. J. (2017). Predict- ing human similarity judgments with distributional models: The value of word associations.Proceed- ings of the Twenty-Sixth International Joint Confer- ence on Artificial Intelligence, 4806–4810. https:// doi.org/10.24963/ijcai.2017/671 Dubossarsky, H., ...

work page doi:10.3758/s13428-018-1115-7 2017

[3] [3]

(2018, March).Learning word vectors for 157 languages

https://doi.org/10.3758/BF03211553 16AESCHBACH Grave, E., Bojanowski, P., Gupta, P., Joulin, A., & Mikolov, T. (2018, March).Learning word vectors for 157 languages. arXiv. https://doi.org/10.48550/arXiv. 1802.06893 Gurevych, I. (2005). Using the structure of a concep- tual network in computing semantic relatedness. In R. Dale, K.-F. Wong, J. Su, & O. Y ....

work page doi:10.3758/bf03211553 2018

[4] [4]

Computer Graphics Forum 31, 2pt2 (2012), 519–528

https://doi.org/10.1111/j.1467- 9280.2009. 02365.x hunspell. (2022).Hunspell. (Version 1.7.2) [Source code]. Github. https://github.com/hunspell/hunspell Hussain, Z., Binz, M., Mata, R., & Wulff, D. U. (2024). A tutorial on open-source large language models for behavioral science.Behavior Research Methods, 56(8), 8214–8237. https://doi.org/10.3758/s13428-...

work page doi:10.1111/j.1467- 2009

[5] [5]

unknown word

https://doi.org/10.1037/0033-2909.87.2.245 Steyvers, M., & Tenenbaum, J. B. (2005). The large-scale structure of semantic networks: Statistical analy- ses and a model of semantic growth.Cognitive Sci- ence,29(1), 41–78. https :/ /doi . org/10 . 1207/ s15516709cog2901_3 The Stanford Natural Language Processing Group. (n.d.). Stanford log-linear part-of-spe...

work page doi:10.1037/0033-2909.87.2.245 2005

[6] [6]

Beispiele: Cue: Mensch; Response: Homo sapiens; Korrigiert: Homo sapiens Cue: Theorie; Response: The Big Bang Theory; Korrigiert: The Big Bang Theory

Wenn eine Response a) richtig geschrieben ist auf Deutsch, b) richtig geschrieben ist in einer anderen Sprache und kein typisches, deutsches falsch geschriebenes Wort ist, oder c) ein korrekt geschriebener Eigenname ist: Die originale Response verwenden. Beispiele: Cue: Mensch; Response: Homo sapiens; Korrigiert: Homo sapiens Cue: Theorie; Response: The B...

work page

[7] [7]

Beispiele: Cue: Argument; Response: Diskussoin; Korrigiert: Diskussion Cue: Marketing; Response: Werbungh; Korrigiert: Werbung

Wenn eine Response falsch geschrieben ist und (in Betracht des Cues und der anderen Assoziationen) die richtige Schreibweise zugerordnet werden kann: Die korrekt geschriebene Response verwenden. Beispiele: Cue: Argument; Response: Diskussoin; Korrigiert: Diskussion Cue: Marketing; Response: Werbungh; Korrigiert: Werbung

work page

[8] [8]

Beispiele: Cue: hören; Response: sehen, fühlen, riechen; Korrigiert: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Korrigiert: feinkörnig

Wenn eine Response Wortkonstrukte enthält die auf mehrere Responses in einem Antwortfeld hinweisen: Die erste Response verwenden. Beispiele: Cue: hören; Response: sehen, fühlen, riechen; Korrigiert: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Korrigiert: feinkörnig

work page

[9] [9]

Beispiele: Cue: lokal; Response: Lokal/Bar; Korrigiert: Bar

Wenn eine Assoziation Wortkonstrukte enthält, die den Cue wiederholen und eine zusätzliche, eigenständige Komponente enthalten: Die korrekt geschriebene eigenständige Komponente verwenden. Beispiele: Cue: lokal; Response: Lokal/Bar; Korrigiert: Bar

work page

[10] [10]

Wenn eine Response ein unvollständiges Wort ist aber in Kombination mit dem Cue Sinn macht: Die sinnvolle Kombination verwenden. Beispiele: Cue: Gebiet; Response: Hohheits; Korrigiert: Hoheitsgebiet Cue: notwendig; Response: keit; Korrigiert: Notwendigkeit English translation of the system prompt: You help me correct potentially misspelled responses in a ...

work page

[11] [11]

Examples: Cue: Mensch; Response: Homo sapiens; Corrected: Homo sapiens Cue: Theorie; Response: The Big Bang Theory; Corrected: The Big Bang Theory

If a response a) is correctly spelled in German, b) is correctly spelled in another language and is not a typical, misspelled German word, or c) is a correctly spelled proper name: Use the original response. Examples: Cue: Mensch; Response: Homo sapiens; Corrected: Homo sapiens Cue: Theorie; Response: The Big Bang Theory; Corrected: The Big Bang Theory

work page

[12] [12]

SWOW-DE21 Examples: Cue: Argument; Response: Diskussoin; Corrected: Diskussion Cue: Marketing; Response: Werbungh; Corrected: Werbung

If a response is misspelled and (considering the cue and the other associations) the correct spelling can be assigned: Use the correctly spelled response. SWOW-DE21 Examples: Cue: Argument; Response: Diskussoin; Corrected: Diskussion Cue: Marketing; Response: Werbungh; Corrected: Werbung

work page

[13] [13]

Examples: Cue: hören; Response: sehen, fühlen, riechen; Corrected: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Corrected: feinkörnig

If a response contains word constructs that indicate multiple responses in one answer field: Use the first response. Examples: Cue: hören; Response: sehen, fühlen, riechen; Corrected: sehen Cue: Schleifpapier; Response: feinkörnig/grobkörnig; Corrected: feinkörnig

work page

[14] [14]

Examples: Cue: lokal; Response: Lokal/Bar; Corrected: Bar

If an association contains word constructs that repeat the cue and contain an additional, independent component: Use the correctly spelled independent component. Examples: Cue: lokal; Response: Lokal/Bar; Corrected: Bar

work page

[15] [15]

If a response is an incomplete word but makes sense in combination with the cue: Use the meaningful combination. Examples: Cue: Gebiet; Response: Hohheits; Corrected: Hoheitsgebiet Cue: notwendig; Response: keit; Corrected: Notwendigkeit User Prompt For each incorrectly spelled response, the following prompt template was used to obtain a correction. Respo...

work page 2020