Psycholinguistic Norms Reference¶

Complete documentation of all word properties available in PhonoLex.

Overview¶

PhonoLex ships ~150 word-property columns derived from a mix of CMU-computed values, PhonoLex in-house gpt-4.1-mini cloze-prompt ratings, and PhonoLex-derived corpus frequencies (from FineWeb-Edu). The original-author papers cited per property below are scale anchors for the in-house derivations, not redistribution sources — the values served by PhonoLex are PhonoLex's, validated against held-out oracles where applicable.

Categories surfaced in the API (/api/property-metadata) and the SLP-curated UI accordions:

Phonological Complexity — syllable_count, phoneme_count, wcm_score, cv_shape
Phonotactic Probability — phono_prob_avg, positional_prob_avg, stressed variants, neighborhood_density
Lexical Frequency — frequency, log_frequency, contextual_diversity (FineWeb-Edu); CYP-LEX child-corpus bands
Lexical Timing — aoa (PhonoLex in-house gpt-4.1-mini cloze; 1-7 age-banded scale, Spearman 0.868 vs Glasgow)
Semantic — concreteness, familiarity, boi, iconicity, socialness, semantic_diversity (PhonoLex in-house)
Affective — valence, arousal (PhonoLex in-house, Warriner-scale)
Morphological — morpheme_count, n_prefixes, n_suffixes, is_monomorphemic (in-house + MorphyNet)

Lexicon: ~125,000 CMU-phonology entries; the 47,915-word canonical content-word vocabulary (selected by content-POS frequency mass plus a WordNet lemma-realness check that excludes proper-noun instances, with a curated deny/allow list on top) carries the full norm set. Non-canonical entries (PROPN / PRON / function words) carry phonology plus corpus-derived frequency and semantic-diversity statistics where covered, but not the in-house psycholinguistic norms.

Coverage notes:

Frequency-class columns (freq_*) treat value=0 as NULL when computing percentiles, so zero-occurrence words don't cluster at a misleading mid-rank.
Percentile bounds in the route layer use NULL-fail semantics (a word lacking the percentile fails the bound); raw norm bounds use NULL-pass semantics.

Retired columns (don't expect to find them in current responses): dominance (Warriner D axis not re-derived in the in-house build), prevalence, aoa_kuperman, elp_lexical_decision_rt, Lancaster sensorimotor channels (auditory, visual, haptic, gustatory, olfactory, interoceptive, hand_arm, foot_leg, head, mouth, torso), size, freq_age_adult, and — removed 2026-07-12 (PHON-161) — the TalkBank developmental frequencies (freq_age_2y/5y/8y/12y, freq_age_all, and the raw *_pb_* / *_childes_* band columns).

Phonological Complexity (4 Properties)¶

Syllables¶

Source: Syllabification algorithm based on English phonotactic constraints

Range: 1-5 syllables

Coverage: ~100% of the full ~125K lexicon

Description: Number of syllables in the word, determined by syllabification algorithm using maximal onset principle and sonority sequencing.

Algorithm: 1. Identify vowel nuclei (all vowels and syllabic consonants) 2. Assign consonants to syllables using maximal onset principle 3. Apply English phonotactic constraints (legal clusters, sonority) 4. Count resulting syllables

Examples: - 1 syllable: cat, dog, strength, spraitz - 2 syllables: happy, table, window, around - 3 syllables: computer, banana, elephant - 4 syllables: university, information - 5 syllables: congratulations, administrative

Clinical use: Early intervention typically targets monosyllabic words. Multisyllabic words added as complexity increases.

Research use: Syllable count correlates with word duration, phonological complexity, and processing time.

Phonemes¶

Source: CMU Pronouncing Dictionary (ARPAbet converted to IPA)

Range: 1-10+ phonemes

Coverage: ~100% of the full ~125K lexicon

Description: Number of phoneme segments in the IPA transcription. Diphthongs count as single phonemes (e.g., /aɪ/ in "time").

Counting rules: - Each IPA symbol = 1 phoneme - Diphthongs (/aɪ/, /aʊ/, /ɔɪ/, /oʊ/, /eɪ/) = 1 phoneme - Consonant clusters (e.g., /str/) count each phoneme separately (3 phonemes) - Affricates (/tʃ/, /dʒ/) = 1 phoneme each

Examples: - 1 phoneme: a /ə/, I /aɪ/ - 2 phonemes: at /æt/, go /goʊ/ - 3 phonemes: cat /kæt/, dog /dɔg/ - 4 phonemes: spray /spreɪ/, think /θɪŋk/ - 5+ phonemes: strength /strɛŋkθ/ (7 phonemes)

Clinical use: Simple words typically have ≤4 phonemes. Higher phoneme counts increase memory load and articulatory complexity.

Research use: Phoneme count correlates with word length, complexity, and neighborhood density.

Note: Phoneme count is NOT the same as letter count. "through" has 3 phonemes (/θru/) but 7 letters.

WCM (Word Complexity Measure)¶

Source: Stoel-Gammon (2010)

Range: 0-15 (theoretical maximum higher for very complex words)

Coverage: ~100% of the full ~125K lexicon

Description: Composite measure of phonological complexity based on 8 parameters reflecting developmental phonology and articulatory difficulty.

Algorithm (8 parameters):

More than 2 syllables: +1
Applies to words with 3+ syllables
Example: "elephant" (3 syllables) → +1
Non-initial stress: +1
Applies when primary stress is NOT on first syllable
Example: "banana" (stress on 2nd syllable) → +1
Word-final consonant: +1
Applies to all words ending in a consonant
Example: "cat" /kæt/ → +1
Consonant cluster: +1 per cluster
Cluster = 2+ adjacent consonants in same syllable
Example: "spray" /spreɪ/ has cluster /spr/ → +1
Example: "strength" /strɛŋkθ/ has clusters /str/ and /ŋkθ/ → +2
Velar: +1 per occurrence
Velars: /k/, /g/, /ŋ/
Example: "king" /kɪŋ/ has /k/ and /ŋ/ → +2
Liquid/Rhotic: +1 per occurrence
Liquids/Rhotics: /l/, /ɹ/
Example: "real" /ɹil/ has /ɹ/ and /l/ → +2
Fricative/Affricate: +1 per occurrence
Fricatives: /f/, /v/, /θ/, /ð/, /s/, /z/, /ʃ/, /ʒ/, /h/
Affricates: /tʃ/, /dʒ/
Example: "fish" /fɪʃ/ has /f/ and /ʃ/ → +2
Voiced fricative/affricate: +1 additional per occurrence
Voiced fricatives: /v/, /ð/, /z/, /ʒ/
Voiced affricates: /dʒ/
Example: "zoo" /zu/ has /z/ → +1 (fricative) +1 (voiced) = +2 total for /z/

Worked example: "cat" /kæt/

1. More than 2 syllables: 0 (1 syllable)
2. Non-initial stress: 0 (stress IS initial)
3. Word-final consonant: +1 (ends in /t/)
4. Consonant clusters: 0 (no clusters)
5. Velars: +1 (/k/)
6. Liquids/rhotics: 0 (none)
7. Fricatives/affricates: 0 (none)
8. Voiced fricatives: 0 (none)

Total WCM: 2

Interpretation: - 0-3: Simple words (cat, dog, bed) - 4-6: Moderate complexity (spray, think, snake) - 7-10: High complexity (splash, strength, squirrel) - 11+: Very high complexity (strengths, splashed)

Clinical use: WCM correlates with age of acquisition and production accuracy in children. Studies typically used simple words (WCM ≤3) for early intervention.

Research use: WCM provides quantitative measure of phonological complexity for stimulus matching and developmental analysis.

References: - Stoel-Gammon, C. (2010). The Word Complexity Measure: Description and application to developmental phonology and disorders. Clinical Linguistics & Phonetics, 24(4-5), 271-282.

Phonotactic Probability (6 Properties)¶

PhonoLex computes six phonotactic properties per word using the Vitevitch & Luce (2004) method, computed directly from the CMU Pronouncing Dictionary: unstressed and stress-marked variants of biphone probability, positional segment probability, and neighborhood density. The two unstressed averages are detailed below. str_phono_prob_avg and str_positional_prob_avg use the same methods over stress-marked phoneme sequences; neighborhood_density and str_neighborhood_density count phonological neighbors at edit distance 1 (unstressed and stressed, respectively) rather than averaging a probability — see the Data Coverage Summary table below for the full column list.

Biphone Probability (Average)¶

Source: Method from Vitevitch & Luce (2004); PhonoLex computes the values directly from the full CMU Pronouncing Dictionary (~134K words)

Range: 0-1 (continuous)

Coverage: ~100% of the full ~125K lexicon

Description: Mean biphone probability across all phoneme pairs in the word. Higher values indicate more typical, phonotactically "legal" sound sequences in English.

What it measures: The probability of phoneme sequences (biphones) occurring in English words, averaged across all biphones in the word.

Algorithm: 1. Syllabify word into onset-nucleus-coda structures 2. Extract all biphone transitions: - Within onset (e.g., /sp/ in "spray") - Onset-to-nucleus (e.g., /s/-/ɪ/ in "sit") - Nucleus-to-coda (e.g., /æ/-/t/ in "cat") - Within coda (e.g., /st/ in "fast") 3. Calculate probability of each biphone from full CMU corpus 4. Average probabilities across all biphones in the word

Worked example: "cat" /kæt/

Syllable: /kæt/
 Onset: /k/
 Nucleus: /æ/
 Coda: /t/

Biphone transitions:
 1. /k/ → /æ/ (onset-to-nucleus): P = 0.0823
 2. /æ/ → /t/ (nucleus-to-coda): P = 0.0412

Average biphone probability: (0.0823 + 0.0412) / 2 = 0.0618

Interpretation: - 0.00-0.02: Very low probability (unusual sound sequences) - "strengths", "twelfths" - 0.02-0.05: Low-moderate probability - "splash", "squid" - 0.05-0.10: Moderate-high probability - "cat", "dog", "jump" - 0.10+: Very high probability (very typical sequences) - "mama", "no", "see"

Clinical use: Phonotactic probability correlates with: - Word learning rate (high probability = faster learning) - Production accuracy (high probability = more accurate) - Neighborhood density effects (high probability words have denser neighborhoods)

Research use: Phonotactic probability is key for: - Word learning studies (probability facilitates acquisition) - Speech perception (high probability aids recognition) - Phonological development (children acquire high-probability patterns first)

References: - Vitevitch, M. S., & Luce, P. A. (2004). A Web-based interface to calculate phonotactic probability for words and nonwords in English. Behavior Research Methods, Instruments, & Computers, 36(3), 481-487. (Method origin; PhonoLex computes the value directly from the CMU Pronouncing Dictionary.)

Positional Segment Probability (Average)¶

Source: Method from Vitevitch & Luce (2004); computed in PhonoLex from the CMU Pronouncing Dictionary

Range: 0-1 (continuous)

Coverage: ~100% of the full ~125K lexicon

Description: Mean probability of individual phonemes occurring at their position in the word, averaged across all phonemes in the word.

What it measures: How typical each phoneme is at its position index, independent of sequence probabilities.

Position means position in the phoneme string, not syllable role. Position 0 is the first phoneme of the word, position 1 the second, and so on. The implementation (packages/data/src/phonolex_data/phonology/phonotactics.py, _compute_positional_stats) enumerates the flat phoneme list and never consults syllable structure. Earlier versions of this page described an onset/nucleus/coda calculation, including a worked example with invented probabilities; no such computation exists. For a monosyllable like cat the two readings coincide, which is how the error survived — they diverge as soon as a word has more than one syllable.

Algorithm: 1. For each phoneme in the word, take its index in the phoneme string 2. Calculate the probability of that phoneme at that index across the full CMU lexicon 3. Average across all phonemes in the word

Worked example: "cat" /kæt/ (computed from the shipped 125,756-word CMU lexicon)

Positional probabilities:
 1. /k/ at index 0: P = 0.0977 (12,291 of 125,756 word-initial phonemes)
 2. /æ/ at index 1: P = 0.0763 (9,587 of 125,715 phonemes at index 1)
 3. /t/ at index 2: P = 0.0590 (7,358 of 124,782 phonemes at index 2)

Average positional probability: (0.0977 + 0.0763 + 0.0590) / 3 = 0.0777

The index totals shrink across positions because shorter words run out of phonemes — only 124,782 words have a third phoneme at all.

Comparison with biphone probability: - Biphone probability: measures phoneme sequences (transitions between adjacent phonemes) - Positional probability: measures individual phoneme frequencies at a position index

Interpretation: - 0.00-0.02: rare phoneme at that position - 0.02-0.05: uncommon - 0.05-0.10: common - 0.10+: very common

Clinical use: Positional probability can guide phoneme selection — a high value means the phoneme occurs frequently at that point in a word, which is useful when picking common sound targets.

Research use: Positional probability isolates segment frequency effects from sequence effects, useful for teasing apart different influences on word processing.

Lexical Properties (2 Properties)¶

Frequency¶

Source: PhonoLex in-house derivation from FineWeb-Edu (~800M tokens, ~1M docs). License: ODC-BY 1.0.

Range: 0-1000+ (per million words, continuous)

Coverage: ~100% of the 47,915-word canonical vocabulary; corpus frequency also covers ~57K non-canonical entries.

Description: Word frequency computed in-house from the FineWeb-Edu corpus (~800M tokens across ~1M documents of curated educational web text). PhonoLex counts tokens (with a spaCy POS pass) and normalizes to occurrences per million words.

Data collection: PhonoLex derivation from HuggingFace FineWeb-Edu (ODC-BY 1.0). Note the corpus's educational-register skew: academic/instructional vocabulary ranks somewhat higher than it would in conversational speech.

Units: Occurrences per million words (raw frequency, not log-transformed in database).

Interpretation:

Range	Label	Examples	Notes
0-1	Extremely rare	flabbergast, obfuscate, pusillanimous	May be technical or archaic
1-5	Very rare	whimsical, erstwhile, penchant	Low-frequency vocabulary
5-20	Uncommon	mansion, skeptical, glimpse	Moderately educated vocabulary
20-100	Common	happy, table, question, important	Everyday vocabulary
100-500	Very common	good, people, know, think	Core vocabulary
500+	Extremely common	the, a, to, of, and, I, you	Function words + core content

Distribution: Highly skewed. Most words have frequency < 10. Top 100 words account for ~50% of all tokens.

Clinical use: Studies typically used high-frequency words (> 20) for functional vocabulary. Low-frequency words may be unfamiliar even to adults.

Research use: Frequency is the strongest predictor of word recognition speed, naming accuracy, and age of acquisition. Essential control variable for psycholinguistic studies.

Why FineWeb-Edu: - Large, recent corpus (~800M tokens) with a clean redistribution-friendly license (ODC-BY 1.0) - In-house derivation — PhonoLex owns the per-word statistics rather than redistributing a third-party norm set - Companion columns from the same pass: log_frequency (log-Zipf normalization) and contextual_diversity (document-level diversity)

References: - Penedo, G., Kydlíček, H., Lozhkov, A., et al. (2024). FineWeb-Edu: an open and high-quality dataset for educational content. License: ODC-BY 1.0.

Age of Acquisition (AoA)¶

Source: PhonoLex in-house derivation: gpt-4.1-mini cloze with logprob expected-value extraction over a 1-7 scale where each step is anchored to an age band (1 = 0-2 yrs, 2 = 3-4, 3 = 5-6, 4 = 7-8, 5 = 9-10, 6 = 11-12, 7 = 13 yrs+). Anchor examples drawn from the published Glasgow + Kuperman protocols. Validated against held-out human norms.

Range: 1-7 (continuous via logprob expected-value)

Validation: Spearman 0.868 vs Glasgow Norms (N=5,551, CC BY 4.0 oracle, kept locally as data/norms/_oracles/GlasgowNorms.xlsx); Pearson 0.816 vs Kuperman 2012 on N=500 Glasgow-unseen rows (sanity oracle only — not redistributed).

Coverage: ~100% of canonical (47,915 rows).

Rating scale (model prompt anchors):

Value	Age band	Examples (high-confidence)
1	0-2 yrs	mum, mama, dad, daddy, ball
2	3-4 yrs	cat, dog, baby, happy
3	5-6 yrs	read, school, friend, story
4	7-8 yrs	science, history, multiply
5	9-10 yrs	democracy, equation, evaporate
6	11-12 yrs	hypothesis, analyze, philosophical
7	13 yrs+	subpoena, oligarchy, tariff, epistemology

Distribution (production seed, integer-binned): values cluster around 4-5 with light tails at the extremes — expected for a cloze derivation over a 7-anchor age-band scale, since most general-purpose English vocabulary is acquired in mid-elementary years.

Predictive validity: AoA predicts word recognition speed and naming accuracy beyond frequency effects. Earlier-acquired words are processed faster even when frequency is matched.

Clinical use: Match target/comparison words on AoA for developmental appropriateness. Early intervention typically uses AoA ≤ 3, later therapy uses AoA 3-5.

Research use: AoA is critical for: - Developmental studies (ensuring age-appropriate vocabulary) - Semantic processing research (earlier words = stronger semantic networks) - Language disorders (children with SSD/DLD show delayed AoA)

Limitation: Subjective ratings may not perfectly reflect actual acquisition age. Cultural and educational differences affect ratings.

References: - Scott, G. G., Keitel, A., Becirspahic, M., Yao, B., & Sereno, S. C. (2019). The Glasgow Norms: Ratings of 5,500 words on nine scales. Behavior Research Methods, 51, 1258-1270. - Kuperman, V., Stadthagen-Gonzalez, H., & Brysbaert, M. (2012). Age-of-acquisition ratings for 30,000 English words. Behavior Research Methods, 44, 978-990.

Semantic Properties (6 Properties)¶

Familiarity¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Glasgow Norms / Scott et al. (2019) is the scale anchor)

Range: 1-7 (Likert scale)

Coverage: ~100% of the canonical vocabulary

Description: Subjective ratings of how familiar the word is to the rater. 1 = very unfamiliar, 7 = very familiar.

Rating scale:

Value	Description	Examples
1-2	Very unfamiliar	pusillanimous, obstreperous, sesquipedalian
3-4	Moderately unfamiliar	erstwhile, whimsical, penchant
5-6	Moderately familiar	analyze, determine, significant
6-7	Very familiar	cat, happy, run, good, see, make

What it measures: Subjective experience of word knowledge, independent of actual usage frequency.

Distinction from frequency: Familiarity ≠ frequency: - "elephant" = high familiarity, moderate frequency (rarely used but well-known) - "pursuant" = low familiarity, moderate frequency (legal jargon, used often in specific contexts)

Correlation with frequency: ~0.65 correlation. Frequency is objective (corpus counts), familiarity is subjective (personal experience).

Collection method: PhonoLex in-house gpt-4.1-mini cloze derivation (logprob expected-value over the 7-point scale), with the human-rated Glasgow Norms FAM column as the validation anchor.

Predictive validity: Familiarity predicts lexical decision speed BEYOND frequency. Familiar words recognized faster even when frequency matched.

Clinical use: High-familiarity words are typically targeted first: - More accessible in therapy - Better generalization - Functional for daily communication

Research use: Familiarity useful for: - Controlling subjective knowledge vs. objective usage - Understanding individual differences (vocabulary size, education) - Semantic memory research

Limitation: Familiarity ratings vary more across individuals than frequency/imageability. Participants' vocabulary size and education affect ratings.

References: - Scott, G. G., et al. (2019). The Glasgow Norms: Ratings of 5,500 words. Behavior Research Methods, 51, 1258-1270.

Concreteness¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Brysbaert et al. (2014) is the scale anchor; validated against held-out Brysbaert oracle)

Range: 1-5 (Likert scale)

Coverage: ~100% of the canonical vocabulary

Description: Rated degree to which a word refers to something perceptible by the senses. 1 = very abstract, 5 = very concrete.

Rating scale:

Value	Description	Examples
1-2	Very abstract	truth, love, democracy, significance, concept
2-3	Moderately abstract	think, believe, important, consider
3-4	Moderately concrete	read, walk, happy, eat, make
4-5	Very concrete	cat, tree, table, water, red, apple

What it measures: Physical, tangible referents vs. abstract concepts. NOT the same as imageability: - "running" = moderate concreteness (action), high imageability (easy to imagine) - "table" = high concreteness (object), high imageability (easy to imagine)

Collection method: PhonoLex in-house gpt-4.1-mini cloze derivation (logprob expected-value over the 5-point Brysbaert scale), validated against the held-out human-rated Brysbaert et al. (2014) norms as oracle.

Concrete-Abstract continuum: - Concrete: Objects (table, cat), actions (run, jump), perceptual properties (red, loud) - Abstract: Emotions (love, anger), concepts (truth, democracy), mental states (think, believe)

Correlation with imageability: ~0.85, but NOT identical: - Concrete nouns: high concreteness, high imageability (cat, tree) - Actions: moderate concreteness, high imageability (running, jumping) - Abstract nouns: low concreteness, low imageability (truth, democracy)

Predictive validity: Concreteness predicts: - Naming speed (concrete > abstract) - Semantic processing (concrete = faster, more automatic) - Memory (concrete = better recall) - Aphasia severity (concrete words spared longer)

Concreteness effect: Across many tasks, concrete words are processed faster and more accurately than abstract words.

Clinical use: Studies typically use concrete words for: - Early vocabulary intervention - Aphasia therapy (concrete words more accessible) - Semantic therapy (easier to demonstrate and explain)

Research use: Concreteness is key for: - Semantic memory research (concrete vs. abstract processing) - Aphasia studies (concrete-abstract dissociation) - Embodied cognition (concrete words = sensory-motor grounding)

References: - Brysbaert, M., Warriner, A. B., & Kuperman, V. (2014). Concreteness ratings for 40 thousand generally known English word lemmas. Behavior Research Methods, 46, 904-911.

Body-Object Interaction (BOI)¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Pexman et al. (2019) is the scale anchor)

Range: 1-7 (Likert scale)

Coverage: ~100% of the canonical vocabulary

Description: Ease of physical interaction with the word's referent. Higher values = easier bodily interaction (e.g., "cup," "ball"); lower values = referents the body cannot readily act on (e.g., "cloud," "justice"). BOI captures sensorimotor experience with objects and predicts processing advantages for high-BOI words in lexical and semantic tasks.

References: - Pexman, P. M., Muraki, E., Sidhu, D. M., Siakaluk, P. D., & Yap, M. J. (2019). Quantifying sensorimotor experience: Body-object interaction ratings for more than 9,000 English words. Behavior Research Methods, 51, 453-466.

Iconicity¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Winter et al. (2024) is the scale anchor)

Range: 1-7

Coverage: ~100% of the canonical vocabulary

Description: How much the word's sound resembles its meaning. Higher values = stronger sound-meaning resemblance (e.g., onomatopoeia like "buzz," "pop"); lower values = arbitrary form-meaning mappings. Iconicity is of particular interest in early word learning and sound-symbolism research.

References: - Winter, B., Lupyan, G., Perry, L. K., Dingemanse, M., & Perlman, M. (2024). Iconicity ratings for 14,000+ English words. Behavior Research Methods, 56, 1640-1655.

Socialness¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Diveica et al. (2023) is the scale anchor)

Range: 1-7 (Likert scale)

Description: Degree to which the word involves people or social interaction. Higher values = more socially relevant (e.g., "friend," "gossip"); lower values = socially neutral referents. Not currently surfaced as a filter in the app UI.

References: - Diveica, V., Pexman, P. M., & Binney, R. J. (2023). Quantifying social semantics: An inclusive definition of socialness and ratings for 8,388 English words. Behavior Research Methods, 55, 461-473.

Semantic Diversity¶

Source: PhonoLex in-house embedding-derived statistics (Qwen3-Embedding representations + topic clustering over FineWeb-Edu contexts; Hoffman et al. (2013) is the methodological anchor)

Range: 0-5 (headline metric)

Description: How many distinct topical contexts the word appears in. Higher values = the word is used across more distinct topics; lower values = usage is topically concentrated. The headline metric is semd_topic (with semantic_diversity retained as an alias); n_topics_for_word counts the distinct topics directly, and semd_vn / semd_h13 are research-stage diagnostic variants. Semantic diversity indexes contextual ambiguity and predicts word-processing differences beyond frequency.

References: - Hoffman, P., Lambon Ralph, M. A., & Rogers, T. T. (2013). Semantic diversity: A measure of semantic ambiguity based on variability in the contextual usage of words. Behavior Research Methods, 45, 718-730.

Affective Properties (2 Properties)¶

Valence and arousal are PhonoLex in-house derivations on the Warriner et al. (2013) scale; the Warriner norms serve as the scale anchor and held-out validation oracle, not a redistribution source. The Warriner D (dominance) axis was retired and is not re-derived.

Valence¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Warriner et al. (2013) VAD is the scale anchor; Spearman 0.836 vs held-out Warriner oracle on N=500 pilot for valence)

Range: 1-9 (Likert scale)

Coverage: ~100% of the canonical vocabulary

Description: Emotional positivity/negativity of the word. 1 = very negative, 9 = very positive, 5 = neutral.

Rating scale:

Value	Description	Examples
1-3	Very negative	death, hate, war, cancer, torture, failure
3-4	Moderately negative	sad, angry, sick, worried, problem
4-6	Neutral	table, chair, walk, see, book, paper
6-7	Moderately positive	happy, good, friend, smile, successful
7-9	Very positive	love, joy, paradise, excellent, wonderful

What it measures: Affective tone, emotional charge. NOT the same as arousal.

Collection method: PhonoLex in-house gpt-4.1-mini cloze derivation (logprob expected-value over the 9-point Warriner scale), validated against the held-out human-rated Warriner et al. (2013) norms as oracle.

Valence dimensions: - Positive valence: Pleasant, desirable, approach motivation - Negative valence: Unpleasant, aversive, avoidance motivation - Neutral valence: No emotional tone

Independence from arousal: Valence and arousal are orthogonal: - High valence + high arousal: excited, thrilled, joyful - High valence + low arousal: calm, peaceful, relaxed - Low valence + high arousal: angry, terrified, panicked - Low valence + low arousal: sad, depressed, bored

Predictive validity: Valence predicts: - Attention (negative valence = attentional capture) - Memory (emotional valence = better encoding than neutral) - Processing speed (extreme valence = slower processing than neutral)

Clinical use: Affective vocabulary useful for: - Social-emotional language therapy - Perspective-taking (understanding others' emotions) - Narrative therapy (emotional content in stories)

Research use: Valence is key for: - Emotion processing research - Mood disorders (depression = negative valence bias) - Decision-making (valence influences choices)

References: - Warriner, A. B., Kuperman, V., & Brysbaert, M. (2013). Norms of valence, arousal, and dominance for 13,915 English lemmas. Behavior Research Methods, 45, 1191-1207.

Arousal¶

Source: PhonoLex in-house derivation (gpt-4.1-mini cloze-prompt; Warriner et al. (2013) VAD is the scale anchor; Spearman 0.836 vs held-out Warriner oracle on N=500 pilot for valence)

Range: 1-9 (Likert scale)

Coverage: ~100% of the canonical vocabulary

Description: Emotional intensity/activation. 1 = very calm, 9 = very excited/intense, 5 = moderate.

Rating scale:

Value	Description	Examples
1-3	Very low arousal	calm, sleep, quiet, relax, peace
3-4	Moderately low	rest, sit, gentle, soft
4-6	Moderate	walk, think, read, see, talk
6-7	Moderately high	excited, surprised, interesting, busy
7-9	Very high arousal	panic, rage, thrill, ecstatic, terrified

What it measures: Physiological activation, emotional intensity. Independent of valence (positive/negative).

Collection method: PhonoLex in-house gpt-4.1-mini cloze derivation (logprob expected-value over the 9-point Warriner scale), validated against the held-out human-rated Warriner et al. (2013) norms as oracle.

Arousal dimensions: - High arousal: Activating, intense, energizing (excited, angry, scared) - Low arousal: Calming, subdued, relaxing (calm, bored, tired)

Circumplex model (Russell, 1980):

High Arousal
 |
 excited angry
 |
Positive ——— Neutral ——— Negative (Valence)
 |
 calm sad
 |
Low Arousal

Independence from valence: Arousal and valence are orthogonal: - Positive + high arousal: excited, happy, thrilled - Positive + low arousal: calm, peaceful, content - Negative + high arousal: angry, terrified, anxious - Negative + low arousal: sad, depressed, bored

Predictive validity: Arousal predicts: - Attention (high arousal = enhanced attention) - Memory (high arousal = better encoding via amygdala activation) - Processing speed (high arousal = faster/slower depending on task) - Physiological response (high arousal = increased heart rate, skin conductance)

Clinical use: Arousal vocabulary useful for: - Emotional regulation therapy - Anxiety management (identifying high-arousal states) - Social-emotional language

Research use: Arousal is key for: - Emotion research (circumplex model, dimensional theories) - Memory (arousal enhances encoding) - Attention (high arousal captures attention) - Psychophysiology (arousal = ANS activation)

References: - Warriner, A. B., et al. (2013). Norms of valence, arousal, and dominance. Behavior Research Methods, 45, 1191-1207. - Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39(6), 1161-1178.

Data Coverage Summary¶

Coverage is essentially the same across most norm columns because the in-house PhonoLex derivations cover the entire 47,915-word canonical vocabulary uniformly. Frequency-class columns vary by source corpus.

Property Category	Columns	Coverage
Phonological	`syllable_count`, `phoneme_count`, `wcm_score`, `cv_shape`	~100% of the 125K full lexicon
Phonotactic	`phono_prob_avg`, `positional_prob_avg`, `neighborhood_density` + stressed variants	~100% of the 125K full lexicon
Lexical Frequency	`frequency`, `log_frequency`, `contextual_diversity` (FineWeb-Edu)	~90% canonical; ~85% full lexicon
CYP-LEX	`freq_cyplex_7_9/10_12/13`	~80% canonical
Lexical Timing	`aoa`	~100% canonical
Semantic	`concreteness`, `familiarity`, `boi`, `iconicity`, `socialness`, `semantic_diversity`	~100% canonical
Affective	`valence`, `arousal`	~100% canonical
Morphological	`morpheme_count`, `n_prefixes`, `n_suffixes`, `is_monomorphemic`	~100% canonical

Overall: 47,915 canonical content words with full norm coverage; ~125K phonology-bearing entries total. Filtering by a column with NULL coverage excludes the missing rows.

Missing data handling: Words without a property are excluded when filtering by that property in Custom Word Lists tool.

Using Properties in PhonoLex¶

Custom Word Lists¶

Filter words by any combination of properties using AND logic:

Example query:

Pattern: STARTS_WITH /s/
Filter: Frequency ≥ 20
Filter: Syllables = 1
Filter: Concreteness ≥ 4.0
Filter: Valence ≥ 6.0

Result: High-frequency, monosyllabic, highly concrete, positive /s/ words

See Custom Word Lists for complete documentation.

Word Lookup¶

View all available properties for any word in the vocabulary.

See Lookup - Word Lookup for details.

Research Applications¶

Stimulus Control¶

Match experimental conditions on confounding variables:

Phonological: Match on syllables, phonemes, WCM to control phonological complexity

Lexical: Match on frequency and AoA to control familiarity and exposure

Semantic: Match on familiarity and concreteness to control semantic processing

Affective: Match on valence and arousal to control emotional processing

Systematic Manipulation¶

Vary properties of interest while controlling others:

Example 1 - Frequency effect: - High-frequency words (> 100) vs. low-frequency words (< 5) - Matched on: syllables, phonemes, concreteness, valence

Example 2 - Concreteness effect: - Concrete words (concreteness > 4) vs. abstract words (concreteness < 2) - Matched on: frequency, syllables, phonemes, valence

Example 3 - Emotional valence: - Positive words (valence > 7) vs. negative words (valence < 3) - Matched on: frequency, syllables, concreteness, arousal

Clinical Research¶

Evaluate treatment effects while controlling stimulus properties:

Example - Phonological intervention study: - Treatment words: WCM = 6-8, Frequency > 20, AoA < 5 - Control words: WCM = 2-4, Frequency > 20, AoA < 5 - Matched on frequency and AoA, differ on WCM

Psycholinguistic Norms Reference¶

Overview¶

Phonological Complexity (4 Properties)¶

Syllables¶

Phonemes¶

WCM (Word Complexity Measure)¶

Phonotactic Probability (6 Properties)¶

Biphone Probability (Average)¶

Positional Segment Probability (Average)¶

Lexical Properties (2 Properties)¶

Frequency¶

Age of Acquisition (AoA)¶

Semantic Properties (6 Properties)¶

Familiarity¶

Concreteness¶

Body-Object Interaction (BOI)¶

Iconicity¶

Socialness¶

Semantic Diversity¶

Affective Properties (2 Properties)¶

Valence¶

Arousal¶

Data Coverage Summary¶

Using Properties in PhonoLex¶

Custom Word Lists¶

Word Lookup¶

Research Applications¶

Stimulus Control¶

Systematic Manipulation¶

Clinical Research¶

See Also¶