GloVe-Word-Embeddings

This is a library to quickly clean, validate, categorise and embed words for academic research.
Design is based on Wang et al., 2026 (Nature Human Behaviour) and Olson et al., 2021 (PNAS).

This package has five parts:

1. pre — clean, normalise or lemmatize words
2. val — validate against Olson’s list or WordNet nouns
3. cat — classify to proper-noun buckets or WordNet categories
4. mod — load embeddings and turn words or phrases into vectors
5. rat — look up concreteness, complexity, and age-of-acquisition ratings

Embedding models and rating norms are hosted on an AWS S3 bucket and downloaded automatically on first use.
NLTK WordNet and names data are also downloaded automatically on first use of any category helper.

Installation

pip install glove-word-embeddings

Preprocessing (pre)

Clean, normalize or lemmatize raw text before validation, classfication or embedding.

from glove_word_embeddings import pre

pre.return_lowercase("  Cat! ")          # "  cat! "
pre.strip_marks("  Cat! ")               # "  Cat  "
pre.strip_stopword("The Cat")            # "Cat"
pre.strip_space("jar of jam")            # "jarofjam"
pre.remove_stopwords("jar of jam")       # ['jar', 'jam']
pre.remove_stopwords("not just a box", how="heavy") # ['box']
pre.space_check("cat")                   # True   (single token)
pre.space_check("jar of jam")            # False

# Full clean (defaults: lower + drop stopwords + strip marks + strip space)
pre.clean_word("  The Cat! ")            # "cat"
pre.clean_word("jar of jam")             # "jarjam"
pre.clean_word("jar of jam", strip_space_bool=False)  # "jar jam"
pre.clean_word("the", strip_stopword_bool=False)      # "the"

# WordNet lemma (cached)
pre.lemmatize("dogs")                    # "dog"
pre.lemmatize("running", pos="v")        # "run"

Validation (val)

Check words against Olson’s validated list and WordNet.

from glove_word_embeddings import mod, val

val.word("telescope")    # True  (in Olson list, single token)
val.word("jar of jam")   # False (fails space check after cleaning)
val.noun("telescope")    # True  (has a WordNet noun synset)
val.noun("quickly")      # False

m = mod.load("glove-olson-validated")
v = m.vocab_set()
val.vocab("telescope", v)   # True
val.vocab("ice cream", v)   # True if any of "ice cream" / "ice_cream" / "ice-cream" is in v

Categorization (cat)

Categorize against a bucket of proper nouns

Check if words are in certain semantic group, room objects, or pure proper nouns.

from glove_word_embeddings import cat

# Ordinary common word? (ignores WordNet proper-noun instances)
cat.check_common("apple")   # True
cat.check_common("nike")    # False
cat.check_common("aaron")   # False

# Which curated buckets does a word belong to?
cat.check_bucket("dog")                              # {"animals"}
cat.check_bucket("paris", check_common=False)        # {"places"}
cat.check_bucket("apple", check_common=True)         # {"fruits"}  (not brands)

# Count across a list of words
cat.count_bucket(words)                              # any bucket ≥ 5
cat.count_bucket(words, name="environment")          # environment objects
cat.count_bucket(words, name="places", number_of_words=1, check_common=False)
cat.count_bucket(words, name="names",  number_of_words=1, check_common=True)
cat.count_bucket(words, name="brands", number_of_words=1, check_common=True)

check_common=False trusts the seed list (places, environment, semantic groups).
check_common=True drops a seed hit when the word is an ordinary common word (names / brands).

Categorize against WordNet category chains

Return the hypernym ladder of a word (general → specific). The word itself is never included in the list.

from glove_word_embeddings import cat

# Default: longest path of the primary (most frequent) sense
cat.category_chain("dog")
# ['entity', 'physical_entity', 'object', 'whole', 'living_thing',
#  'organism', 'animal', 'chordate', 'vertebrate', 'mammal',
#  'placental', 'carnivore', 'canine']

# Shortest path
cat.category_chain("dog", path="short")
# ['entity', 'physical_entity', 'object', 'whole', 'living_thing',
#  'organism', 'animal', 'domestic_animal']

# Full set of ancestors (union of every path of the primary sense)
cat.category_chain("dog", path="full")
# ['entity', 'physical_entity', 'object', 'whole', 'living_thing',
#  'organism', 'animal', 'chordate', 'domestic_animal', 'vertebrate',
#  'mammal', 'placental', 'carnivore', 'canine']

# Ambiguous words – primary sense only (default)
cat.category_chain("badger", path="full")
# → hierarchy of the “Wisconsin inhabitant” sense

# Ambiguous words – union of *all* senses
cat.category_chain("badger", path="full", sense="union")
# → merges the person hierarchy and the animal hierarchy

Embedding (mod)

Load a vector model and embed single words or short phrases. Embedding models available:

from glove_word_embeddings import mod

m = mod.load("glove-6b-300d")          # downloads on first use, caches locally
m = mod.load("tasa-lsa-300d")
m = mod.load("cbow-baroni-400d")
m = mod.load("cbow-ukwac-subs-300d")
m = mod.load("cbow-subs-300d")
m = mod.load("glove-olson-validated")
m = mod.load("bert-large")

m.embed_exact("cat")                   # exact match only → np.ndarray or None
m.vocab_set()                          # → set of all words in the model

# Multi-word phrases
m.embed_phrase("jar of jam")               # tries exact / underscore / hyphen variants,
                                           # otherwise averages the non-stopword parts
m.embed_phrase("jar of jam", how="heavy")  # remove stopwords as in Beaty & Johnson (2021)

Files are cached in ~/.cache/glove-word-embeddings.

Ratings (rat)

Scalar norms, not embeddings. Load with rat.load, not mod.load. Missing words return None. Stored scales: concreteness is Brysbaert’s 1–5 mean rescaled as (Conc.M − 1) / 4 × 100, complexity is 0–100, age of acquisition is years.

score and score_phrase clean by default (clean=True): lowercase and strip punctuation, using pre.clean_word without dropping stopwords or spaces. Arsenal. hits. ice cream stays a bigram. clean=False is the exact key.

from glove_word_embeddings import rat

r = rat.load("brysbaert-concreteness")   # downloads on first use, caches locally
r.score("dog")                           # 96.25
r.score("ice cream")                     # bigram hit, not an average
r.score("Arsenal.")                      # punctuation stripped, then case-folded
r.score("Arsenal.", clean=False)         # None
r.score_phrase("jar of jam")             # whole phrase, else mean of known tokens
r.score_phrase("jar of jam", how="heavy")
r.vocab_set()

Citations

If you like this package, please cite the relevant works.

Wang et al., (2026) -- Package owner [CITE THIS!]

Wang, D., Huang, D., Shen, H., & Uzzi, B. (2026). A large-scale comparison of divergent creativity in humans and large language models. Nature Human Behaviour, 10(3), 531–540. https://doi.org/10.1038/s41562-025-02331-1

Olson et al., (2021)

Olson, J. A., Nahas, J., Chmoulevitch, D., Cropper, S. J., & Webb, M. E. (2021). Naming unrelated words predicts creativity. Proceedings of the National Academy of Sciences, 118(25), e2022340118. https://doi.org/10.1073/pnas.2022340118

Beaty & Johnson, (2021)

Beaty, R. E., & Johnson, D. R. (2021). Automating creativity assessment with SemDis: An open platform for computing semantic distance. Behavior Research Methods, 53(2), 757–780. https://doi.org/10.3758/s13428-020-01453-w

Johnson et al., (2023)

Johnson, D. R., Kaufman, J. C., Baker, B. S., Patterson, J. D., Barbot, B., Green, A. E., van Hell, J., Kennedy, E., Sullivan, G. F., Taylor, C. L., Ward, T., & Beaty, R. E. (2023). Divergent semantic integration (DSI): Extracting creativity from narratives with distributional semantic modeling. Behavior Research Methods, 55(7), 3726–3759. https://doi.org/10.3758/s13428-022-01986-2

Brysbaert et al., (2014) - Concreteness ratings

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

Kuperman et al., (2012) - Age of acquisition ratings

Kuperman, V., Stadthagen-Gonzalez, H., & Brysbaert, M. (2012). Age-of-acquisition ratings for 30,000 English words. Behavior Research Methods, 44(4), 978–990. https://doi.org/10.3758/s13428-012-0210-4

Maddela et al., (2018) - Word complexity

Maddela, M., & Xu, W. (2018). A word-complexity lexicon and a neural readability ranking model for lexical simplification. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (pp. 3749–3760). Association for Computational Linguistics. https://doi.org/10.18653/v1/D18-1410