nominate

Generate random human names under constraints, weighted by real-world frequency data.

nominate samples forenames and surnames from country-level census/demographic frequency data (derived from the Onomaverse names dataset, CC BY 4.0), so generated names are as common or as rare as you ask them to be. It is built for narrative use ("name this character") but works anywhere you need plausible names.

The full dataset (110 countries, ~6 MB) ships inside the package — no downloads, no network access at runtime.

Features

• Weighted sampling: names are drawn proportionally to how often they actually occur.
• Convention-aware rendering: countries that write the family name first (Vietnam, Korea, China, Hungary, ...) come out that way ("Nguyễn Văn An").
• Constraints: country/region, gender, syllable count, regex, rhyme / phoneme inclusion-exclusion, uniqueness (exclusion lists), and commonness percentiles.
• Partial names: freeze known forenames/surnames and complete the rest.
• Reverse classification: given a name, estimate gender and likely countries.
• Python API and a CLI.

Install

pip install nominate

Quick start (Python)

from nominate import NameGenerator

gen = NameGenerator(countries=["US"], seed=42)

# One forename + one surname, letting the data decide gender:
name = gen.generate()
print(name.full)

# A common-ish two-syllable feminine name starting with A:
name = gen.generate(
    gender="F",
    syllables=(2, 3),
    regex=r"^A",
    percentile=(0.5, 1.0),
)

# Complete a partial name ("I know the first name, need a surname"):
name = gen.generate(forenames=1, surnames=1, frozen_forenames=["James"])

# A sibling for an existing character (shared surname, unique forename):
name = gen.generate(
    frozen_surnames=["Kowalski"],
    exclude={"Anna"},          # already used
)

# Rhymes and phonemes:
name = gen.generate(rhymes_with="Kelly", exclude_phonemes=["TH"])

# Whole regions, or everything:
gen = NameGenerator(region="europe")
gen = NameGenerator()          # all 110 countries

Reverse classification:

from nominate import classify

result = classify("Anna Kowalska")
result.gender_probabilities   # {"F": 0.997, "M": 0.003}
result.countries              # [("PL", 0.75), ("SE", 0.03), ...]

# Family-name-first names are detected and read in the right order:
result = classify("Nguyễn Văn An")
result.forename, result.surname   # ("An", "Nguyễn")
result.surname_first              # True

Quick start (CLI)

# Generate names:
nominate generate -c US -n 10
nominate generate -c US --gender f --syllables 2-3 --regex '^A' \
    --percentile 25-90 -n 5
nominate generate -c US --freeze-forename James -n 3
nominate generate --region europe -n 5
nominate generate -c US --rhymes-with Kelly --exclude-file used_names.txt

# Classify a name:
nominate classify "Anna Kowalska"

# Inspect the bundled data:
nominate data status
nominate data list --region oceania

Constraints reference

Percentiles are within a country and name kind: percentile=(0.9, 1.0) means "the most common 10% of names", (0.0, 0.1) means "the rarest 10%".

By default, gendered sampling trusts the data: a name that is recorded for both genders can be drawn for either, proportionally to its counts. Passing strict_gender=True additionally drops names whose majority gender does not match the requested one.

Data

Per-country Parquet files (given/US.parquet, surname/US.parquet, ...) are bundled at nominate/_bundled/ inside the package. They are produced by the scripts in pipeline/, which merge several sources — the Onomaverse frequency data (via a Hugging Face bucket), the erpel1 Kaggle forenames/surnames dataset, and sigpwned's popular-names-by-country dataset — so that countries with thin census coverage are backfilled rather than silently underrepresented. Supplementary counts are rescaled or Zipf-extrapolated onto each country's measured scale and never override measured data. Everything is enriched with romanizations, ARPAbet phonemes, syllable counts, rhyme keys, and percentile ranks. See pipeline/README.md for details and how to rebuild.

You can point the library at a custom build with NameGenerator(data_dir=...) or the NOMINATE_DATA_DIR environment variable.

Phonetics are Latin-script oriented: names are romanized (via Onomaverse transliterations, falling back to unidecode) and looked up in CMUdict, with a heuristic grapheme-to-phoneme fallback for names CMUdict does not know.

Attribution

Bundled data is derived from:

• Onomaverse names datasets, licensed CC BY 4.0.
• Forenames and Surnames with Gender and Country by erpel1 (Kaggle), licensed Apache 2.0.
• vietnamese-namedb by duyet, licensed MIT.
• popular-names-by-country-dataset by sigpwned, licensed CC0 1.0.

License

MIT (code). Bundled data: CC BY 4.0 (see Attribution).