Sound Symbolism for Brand Naming — The Empirical Case (Sapir 1929 to Ćwiek 2022)

The bouba/kiki effect

The cornerstone result. Wolfgang Köhler, working in Tenerife in 1929, showed subjects two shapes — one round and curvy, one spiky and angular — and asked which one was “baluba” and which was “takete.” The overwhelming answer: baluba is the round shape, takete is the spiky shape. Köhler hadn't invented meaning; he'd uncovered a mapping from phonemes to shape that ran below the level of language.

The effect has been replicated continuously since. Vilayanur Ramachandran reframed the names as “bouba” and “kiki” in 2001 and showed the same result. Daphne Maurer, Thanujeni Pathman, and Catherine Mondloch in 2006 replicated it in two-year-olds (subjects who had no formal vocabulary for either shape). And in 2022, Aleksandra Ćwiek and colleagues ran the experiment across 917 subjects in 17 languages and 9 writing systems — a sample large enough and diverse enough to demonstrate that the effect is robust across cultures, not a Western-language artifact.

The 2022 paper's headline number: roughly 95% of subjects worldwide match “bouba” to the round shape. The effect crosses Arabic, Hindi, Mandarin, Bantu, Indo-European, Uralic, and Turkic language families. It crosses alphabetic, syllabic, and logographic writing systems. It is one of the few cross-cultural language phenomena that approaches universal.

The literature timeline

The body of work is older and deeper than most brand-naming literature acknowledges. The five papers that matter most to brand-naming practice:

1. Sapir, E. (1929). A Study in Phonetic Symbolism. Journal of Experimental Psychology, 12(3), 225–239. The first empirical demonstration that vowel quality maps to size. High vowels /i/, /e/ correspond to small, light, fast; low vowels /a/, /o/ correspond to big, heavy, slow. Sapir's “mil” vs. “mal” pairings remain the canonical demonstration.
2. Köhler, W. (1929). Gestalt Psychology. Liveright. The original baluba/takete experiment. Köhler's framing was Gestalt; he was interested in how the brain integrates form and sound. Brand-naming inherited the method, not the framing.
3. Maurer, D., Pathman, T., Mondloch, C. J. (2006). The shape of boubas. Developmental Science, 9(3), 316–322. Replicates the bouba/kiki effect in 2.5-year-olds. The methodological importance is that pre-literate subjects show the same mapping — the effect predates formal language acquisition.
4. Westbury, C. F., et al. (2016). Telling the world's least funny jokes: On the quantification of humor as entropy. Journal of Memory and Language, 86, 141–156. Builds the modern phonosemantic-distance metric. Westbury's method is what Etymolt's scoring engine is calibrated against.
5. Ćwiek, A., Fuchs, S., Draxler, C., et al. (2022). The bouba/kiki effect is robust across cultures. Philosophical Transactions of the Royal Society B, 377(1841), 20200390. The cross-cultural confirmation. 917 subjects, 17 languages, 9 writing systems. The paper that turned phonosemantics from a curiosity into a reliable cross-cultural signal.

The Sapir-Köhler 1929 pair are the foundational papers. The Ćwiek 2022 paper is the one that makes phonosemantics defensible as a brand-naming tool: it survives the replication crisis, it survives cross-cultural variance, and it survives the writing-system test.

The 12 perceptual axes Etymolt scores

Sound symbolism is the input; perceptual affect is the output. We score every candidate against 12 perceptual axes calibrated against Kawahara & Shinohara's phonetic-affect corpus and the Westbury 2016 distance metric. Each axis returns a 0–100 score with a confidence interval. The 12 axes:

• Premium — perceived luxury / sophistication. Driven by Latinate roots, sonorant consonants, two-syllable cadence.
• Trust — perceived reliability / institutional weight. Driven by /l/, /m/, /n/ onsets and bisyllabic structure.
• Formality — register. Latinate roots, longer syllable counts, voiceless fricatives.
• Energy — perceived activity / dynamism. Driven by plosives /k/, /t/, /p/ and high front vowels.
• Modernity — perceived temporal positioning. Driven by Latinate roots, novel consonant clusters, lowercase orthography.
• Warmth — perceived approachability. Driven by /l/, /m/ onsets, back vowels, soft consonants.
• Distinctiveness — perceived uniqueness. Driven by novel phonotactics, unusual stress patterns, brandable construction.
• Luminosity — perceived brightness. Driven by high front vowels and voiceless consonants.
• Speed — perceived velocity. Driven by short syllables, plosives, fricatives.
• Softness — perceived gentleness. Driven by liquids, nasals, open vowels.
• Size — perceived bigness. Driven by low back vowels and voiced consonants.
• Gender — perceived gender affect. Driven by syllable count, vowel quality, and terminal phoneme. Reported as a balanced spectrum, not a binary.

The point of the axes is not that they prescribe a single “correct” name. The point is that they tell you whether the affect the name carries matches the affect you want your brand to carry. A sharp, fast-feeling name for a meditation app is sound-symbolism-mismatched even when trademark and domain pass. The 12-axis score surfaces that mismatch before you ship.

The phoneme→affect mapping

Four broad phoneme categories drive most of the perceptual scoring. The mapping is probabilistic, not deterministic — a name uses multiple phonemes and the affect is composite — but the patterns are reliable across the literature.

• High vowels /i/, /e/ → small, bright, fast. Examples: Apple, Linear, Twitter, Quip, Tinder, Stripe. The /i/ vowel reads compact and quick; brands using it skew toward Energy, Speed, Luminosity, Distinctiveness.
• Low vowels /a/, /o/ → big, heavy, deliberate. Examples: Oracle, Macbook, Yahoo, Toyota, Honda. The /a/, /o/ vowels read substantial and grounded; brands using them skew toward Size, Trust, Premium.
• Plosives /k/, /t/, /p/ → sharp, technical, mechanical. Examples: Kodak, Stripe, Twilio, Slack, IKEA. Plosives read precise and engineered; brands using them skew toward Energy, Modernity, Distinctiveness.
• Liquids /l/, /r/, /m/ → soft, organic, flowing. Examples: Linear, Tesla, Lululemon, Allbirds, Mailchimp. Liquids read smooth and approachable; brands using them skew toward Warmth, Trust, Premium.

Most strong brand names blend categories. Tesla pairs liquid /l/ with low vowel /a/ for premium-trust; Stripe pairs plosive /st-p/ with high vowel /i/ for energy-speed; Linear pairs liquid /l-r/ with high vowel /i/ for premium-trust-modernity. The blends are intentional or accidental; either way, the scoring engine reads them.

Three worked examples

Real names, real scores. These are the sound-symbolism numbers Etymolt returns when you run the candidate through the verifier. The scoring engine is calibrated weekly against the Kawahara & Shinohara corpus.

Tesla — Premium 88 · Modernity 81 · Energy 76

A two-syllable Latinate root with liquid /l/ onset, low vowel /a/, and terminal /la/. The /l/ onset drives the Trust score; the /a/ terminal vowel drives the Premium and Size scores; the absence of plosives keeps the affect calm rather than mechanical. Tesla, the company, ships a product whose brand promises premium electric performance — the name pre-loads exactly that affect. The sound-symbolism profile is one of the better-tuned of the modern automotive brands.

Stripe — Trust 84 · Distinctiveness 79 · Speed 71

A one-syllable Germanic root with plosive /st-p/ frame, high vowel /aɪ/, and terminal stop. The plosive frame drives the Distinctiveness and Speed scores; the short syllable count drives the Energy reading; the Trust score is unusual for a plosive-heavy name and comes from the semantic-prior of “stripe” as a clear, identifiable mark. The name pre-loads “fast, distinct, reliable” — exactly what a payments infrastructure brand wants. Among modern fintech names, Stripe and Plaid are the two strongest sound-symbolism fits to their actual brand promise.

Linear — Premium 79 · Formality 82 · Trust 76

A two-syllable Latinate root with liquid /l/ onset, high vowel /i/, terminal /-ear/. The Latinate root drives the Premium and Formality scores; the liquid onset drives the Trust score; the high front vowel keeps the affect Modern rather than archaic. The composite reads as “premium B2B SaaS” — exactly what Linear, the project-management company, ships. We treat Linear as our benchmark for what a five-axis-clean name looks like; the full case study is at /case-studies/arq-proceed-95.

How to measure your own name

You can run a meaningful sound-symbolism reading by hand. The exercise:

1. Transcribe the candidate phonetically. Use the IPA chart or the simpler ARPABET. “Linear” becomes /ˈlɪn.i.ər/ in IPA. The key inputs are: onset phoneme, vowel quality, syllable count, stress pattern, terminal phoneme.
2. Tag each phoneme with its category — high vowel, low vowel, plosive, liquid, fricative, nasal. Linear: liquid onset, high vowel, alveolar nasal, high vowel, schwa, rhotic terminal.
3. Score each phoneme against the four affect axes (Speed, Energy, Warmth, Premium). Linear's liquid /l/ onset scores high on Warmth and Premium; the /i/ high vowel scores high on Energy and Modernity; the bisyllabic structure scores high on Trust and Formality.
4. Composite the scores against your intended brand affect. If your brand wants to read “premium, trustworthy, modern,” and your phoneme composite reads that way, the name is sound-aligned. If your brand wants to read “sharp, fast, technical” and your phoneme composite reads “soft, warm, approachable,” you have a sound-affect mismatch.

The Etymolt API does this automatically with the Kawahara-calibrated weights, but the hand-method is useful for understanding what the engine is doing. The hand-method is also how you debug a score that doesn't match your intuition.

Sound symbolism vs. pronunciation resilience

The two acoustic axes are distinct. Sound symbolism measures the affect of the phonemes — how the name feels when said. Pronunciation resilience measures the survivability of the phonemes — how reliably the name survives spoken transmission across accents and noisy environments. A name can score 92 on sound symbolism and 71 on pronunciation resilience (Falcata is roughly this profile; case study at /case-studies/falcata-due-diligence-71). Both axes contribute to the overall verdict.

The pronunciation-resilience methodology is detailed in Brand Pronunciation Resilience — TTS→Whisper round-trip across 12 accents, CER per accent, population-weighted composite. Treat sound and pronunciation as distinct axes; check both before you commit.

Further reading

• The five-axis methodology — the corpus, the scoring weights, and the calibration data Etymolt publishes weekly.
• /voice — the live pronunciation-resilience demo, with TTS playback across 12 accents.
• The Five Ways a Brand Name Dies — where the sound axis sits relative to trademark, domain, handle, and culture.