WordTangible

WordTangible is a simple Python library for analyzing the concreteness and sensory profile of words and text. This can be useful for various natural language processing tasks, readability analysis, stylometry, and linguistic research.

Data is pulled from:

• The Brysbaert dataset (Brysbaert et al., 2014)[1]
• The Muraki multiword-expression dataset (Muraki et al., 2023)[2]
• The Glasgow dataset (Scott et al., 2019)[3]
• The MRC Psycholinguistic Database (Coltheart, 1981; machine-usable dictionary: Wilson, 1988)[4][5]
• The Lancaster Sensorimotor Norms (Lynott et al., 2020)[7] — for the sensory-modality functions

The default rating is a quality-ordered fallback on a 1-5 scale (5 = most concrete): Brysbaert's raw value when a word is in Brysbaert (the largest single-word source, natively 1-5), otherwise Muraki (multiword expressions, same scale and lab lineage), otherwise Glasgow, otherwise MRC, the latter two linearly rescaled to 1-5.

The sources are deliberately not averaged — their normalized distributions have systematically different means, so a linear-rescale average would skew multi-source words rather than reduce noise. This way every value is a real published rating from a single identifiable study (and ~99% of words return Brysbaert's exact published value). The raw per-source ratings, an MRC-free variant, and a mean of the available sources are all accessible via the source parameter (see below).

Features

• Get concreteness ratings for individual words
• Choose the ratings source: the default fallback, any single dataset un-normalized (brysbaert, muraki, glasgow, mrc), an MRC-free variant (open) for commercial use, or a normalized mean
• Rated multiword expressions match as units, longest first: "baseball bat" and "piece of cake" get their own ratings (2,896 Brysbaert compounds + 62,889 Muraki expressions, idioms included) instead of blending their constituent words; disable with match_expressions=False if idiom ratings firing on literal uses ("he ate a piece of cake") is a concern
• Unrated inflected forms score as their WordNet lemma ("whales" as "whale", "replied" as "reply") — worth ~10 points of token coverage on typical fiction; disable with lemma_fallback=False for values strictly comparable to the published norms
• Calculate average concreteness for a given text
• Compute the ratio of concrete to abstract words in a text (with optional add-k smoothing to keep it finite and stable on short texts)
• Report rating coverage — how much of a text the concreteness mean actually rests on
• Sensory-modality analysis via the Lancaster Sensorimotor Norms: per-word perceptual profiles across six modalities (visual, auditory, haptic, interoceptive, gustatory, olfactory), a word's dominant modality, and text-level sensory profiles — how visual vs auditory vs tactile a text is

Installation

You can install WordTangible using pip:

pip install wordtangible

Usage

Here are some basic examples of how to use WordTangible:

from wordtangible import word_concreteness, avg_text_concreteness, concrete_abstract_ratio

# Get concreteness rating for a single word
print(word_concreteness("apple"))  # Output: 5.0 (highly concrete)

# Calculate average concreteness of a text
text = "The abstract concept of love is as tangible as the apple in your hand."
print(avg_text_concreteness(text))  # Output: ~2.9 (mix of concrete and abstract)

# Get the ratio of concrete to abstract words
print(concrete_abstract_ratio(text))  # Output: ~1.0 (balanced concrete and abstract words)

# Smoothed ratio: finite even when a text has no very-abstract words,
# and steadier on short texts (add-k smoothing; default k=0 keeps the
# classic behavior, including float('inf') for the no-abstract case)
print(concrete_abstract_ratio(text, smoothing=1))

# How much of the text the concreteness mean rests on (0.0-1.0):
# a mean at 0.85 coverage is trustworthy; the same mean at 0.12
# (jargon, dialect, names) is noise
from wordtangible import concreteness_coverage
print(concreteness_coverage(text))

Choosing a ratings source

Every function accepts a source parameter:

word_concreteness("apple")               # 5.0   — default fallback, 1-5 scale
word_concreteness("apple", "brysbaert")  # 5.0   — raw Brysbaert, 1-5 scale
word_concreteness("apple", "glasgow")    # 6.824 — raw Glasgow CNC, 1-7 scale
word_concreteness("apple", "mrc")        # 620   — raw MRC CNC, 100-700 scale
word_concreteness("apple", "open")       # 5.0   — like default, but never MRC
word_concreteness("apple", "mean")       # 4.78  — mean of available sources, rescaled to 1-5
word_concreteness("piece of cake")       # 2.8   — Muraki multiword expressions, idioms included

avg_text_concreteness(text, source="open")

# Get the concreteness score and average concreteness for text using exactly Brysbaert and nothing else:
word_concreteness("apple", source="brysbaert", lemma_fallback=False)
avg_text_concreteness(
    text,
    include_stopwords=True,
    source="brysbaert",
    lemma_fallback=False,
)

• default — Brysbaert, else Muraki, else Glasgow, else MRC (the latter two rescaled to 1-5).
• brysbaert / muraki / glasgow / mrc — one dataset's raw, un-normalized values on its native scale; None for words it doesn't rate. Useful for comparing directly against the published norms. If you pass these to concrete_abstract_ratio, adjust its thresholds to the source's scale.
• open — Brysbaert, else Muraki, else Glasgow: excludes the MRC database, whose terms are "for research purposes" (see licensing below), making this the right choice for commercial products.
• mean — the mean of whichever of the four rate the word, each linearly rescaled to 1-5. Beware: linear rescaling doesn't fully align the scales (the sources' normalized means differ systematically), so these values aren't comparable to any single set of published norms.

Sensory-modality analysis

Concreteness says how much a word is tied to experience; the sensorimotor functions say through which senses:

from wordtangible import (
    word_sensorimotor, dominant_modality,
    avg_text_sensorimotor, modality_profile,
)

word_sensorimotor("thunder")
# {"visual": 0.53, "auditory": 3.71, ..., "dominant": "auditory",
#  "max_perceptual": 3.71, "minkowski3_perceptual": ..., "exclusivity": ...}

dominant_modality("perfume")   # "olfactory"
dominant_modality("velvet")    # "haptic"
dominant_modality("nausea")    # "interoceptive"

# Mean 0-5 perceptual strength per modality across a text
avg_text_sensorimotor(text)
# {"visual": 2.69, "auditory": 1.88, "haptic": 0.98, ...}

# Composition: share of rated words dominated by each modality
modality_profile(text)
# {"visual": 0.63, "interoceptive": 0.18, "auditory": 0.17, ...}

These functions accept the same include_stopwords, lemma_fallback, and match_expressions parameters as the concreteness functions (there is no source — the Lancaster norms are the single source). As a taste of what this measures: an 1887 cookbook profiles ~4x higher on gustatory strength than five other reference texts, Austen leads on interoception (inner states), and Darwin is the most visually dominant.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

The WordTangible code is licensed under the MIT License - see the LICENSE file for details.

Data sources & licensing

The bundled ratings file (wordtangible/resources/concreteness_ratings.csv) is derived from third-party datasets, and the MIT license above does not apply to that data. Each source has its own terms:

• Glasgow Norms (Scott et al., 2019): published open access under a Creative Commons Attribution 4.0 license — redistribution of derived data is permitted with attribution, which the citation below provides.
• Brysbaert concreteness norms (Brysbaert et al., 2014): made freely available by the authors (via the Behavior Research Methods supplement and the Ghent CRR lab) and widely redistributed in research software; no formal license accompanies the data, so provenance and citation are provided here as is standard practice.
• Muraki multiword-expression norms (Muraki et al., 2023): distributed by the authors via the paper's OSF repository (osf.io/ksypa), same research-community terms as the Brysbaert norms.
• Lancaster Sensorimotor Norms (Lynott et al., 2020): published open access and distributed via the project's OSF repository (osf.io/7emr6) under a Creative Commons Attribution 4.0 license — redistribution of derived data is permitted with attribution, which the citation below provides.
• MRC Psycholinguistic Database (Coltheart, 1981; Wilson, 1988): the database's distribution terms state that it is available for research purposes. The default ratings fall back to MRC-derived values for the few hundred words none of the other sources rate. If you intend to use WordTangible in a commercial (non-research) product, pass source="open" — it draws only on Brysbaert, Muraki, and Glasgow — or verify the MRC terms for your use case.

The bundled CSV keeps each source's raw rating in its own column, and scripts/build_ratings.py regenerates it from the original datasets (downloaded on demand; the raw files are not stored in this repository).

The lemma fallback additionally uses WordNet (Miller, 1995), which is not bundled either: NLTK downloads it on first use, under the permissive Princeton WordNet license.

This section documents provenance in good faith and is not legal advice.

Citing WordTangible

If you use WordTangible in research, please cite both the tool and the rating datasets your results rest on (for concreteness: all four under the default source; Brysbaert, Muraki, and Glasgow if you use source="open"; the single dataset if you use a raw source — and the Lancaster Sensorimotor Norms[7] if you use the sensory-modality functions). GitHub's "Cite this repository" button generates a citation from CITATION.cff, or use:

Robison, J. (2026). WordTangible (Version 0.7.0) [Computer software]. https://github.com/jrrobison1/wordtangible

@software{robison_wordtangible,
  author  = {Robison, Jason},
  title   = {WordTangible},
  version = {0.7.0},
  year    = {2026},
  url     = {https://github.com/jrrobison1/wordtangible}
}

If your analysis uses the default lemma fallback (lemma_fallback=True), consider also citing WordNet[6], which provides the lemmatization.

Dataset citations are given in full in the References below.

References

[1] Brysbaert, M., Warriner, A. B., & Kuperman, V. (2014). Concreteness ratings for 40 thousand generally known English word lemmas. Behavior Research Methods, 46(3), 904-911. https://doi.org/10.3758/s13428-013-0403-5

[2] Muraki, E. J., Abdalla, S., Brysbaert, M., & Pexman, P. M. (2023). Concreteness ratings for 62,000 English multiword expressions. Behavior Research Methods, 55(5), 2522-2531. https://doi.org/10.3758/s13428-022-01912-6

[3] 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(3), 1258-1270. https://doi.org/10.3758/s13428-018-1099-3

[4] Coltheart, M. (1981). The MRC psycholinguistic database. The Quarterly Journal of Experimental Psychology Section A, 33(4), 497-505. https://doi.org/10.1080/14640748108400805

[5] Wilson, M. (1988). MRC Psycholinguistic Database: Machine-usable dictionary, version 2.00. Behavior Research Methods, Instruments, & Computers, 20(1), 6-10. https://doi.org/10.3758/BF03202594

[6] Miller, G. A. (1995). WordNet: A lexical database for English. Communications of the ACM, 38(11), 39-41. https://doi.org/10.1145/219717.219748 <<<<<<< HEAD

[7] Lynott, D., Connell, L., Brysbaert, M., Brand, J., & Carney, J. (2020). The Lancaster Sensorimotor Norms: Multidimensional measures of perceptual and action strength for 40,000 English words. Behavior Research Methods, 52(3), 1271-1291. https://doi.org/10.3758/s13428-019-01316-z

720131b (Add sensorimotor norms)