Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

numtag

numtag finds every number in a text. For each number it gives the exact words, their character offsets, an exact value, the kind of number and how it is written. It reads number words ("duas dúzias"), digits ("1.234,5") and both together ("2 milhões").

from numtag import tag_numbers

for m in tag_numbers("Comprou duas dúzias de ovos e ganhou um milhão de euros.", "pt-PT"):
    print(m.char_span, repr(m.text), m.value, m.kind.value)
(8, 19) 'duas dúzias' 24 cardinal
(37, 46) 'um milhão' 1000000 cardinal

Install

numtag needs Python 3.11 or later and has no dependencies.

uv pip install --prerelease=allow numtag

To work on the library, install it from a clone with its test extra and run the tests:

uv pip install -e ".[test]"
pytest tests

Quickstart

tag_numbers(text, lang) returns a list of NumberMention objects in reading order. The mentions never overlap.

>>> from numtag import tag_numbers
>>> mention = tag_numbers("The tower is one hundred and twenty-five point five metres tall.", "en")[0]
>>> mention.text, mention.char_span, mention.value
('one hundred and twenty-five point five', (13, 51), '125.5')
>>> mention.kind, mention.form
(<Kind.DECIMAL: 'decimal'>, <Form.WORDS: 'words'>)

The span slices the original text, so text[start:end] is always mention.text. Use the span to replace or mark the number in place.

Values

A value is a canonical digit string, never an int or a float. It has one of three forms.

Form Examples
whole number "24", "-42", "1000000"
decimal "125.5", "0.75"
fraction in lowest terms "1/3", "7/3"

numtag computes each value with string arithmetic on digits, so a value has no length limit and no rounding error. Convert a value when you need a number. int() reads a whole number, decimal.Decimal() reads a whole number or a decimal, and fractions.Fraction() reads all three forms.

>>> from fractions import Fraction
>>> Fraction(tag_numbers("dois e um terço", "pt-PT")[0].value)
Fraction(7, 3)

Read docs/values.md for the full contract.

Languages

lang Lexicon Default scale
pt-BR Brazilian Portuguese short: "bilhão" is 10^9
every other pt tag (pt, pt-PT, pt-AO) European Portuguese long: "bilião" is 10^12
every en tag (en, en-US, en-GB) English short: "billion" is 10^9
el, ar, he, fa, kab Greek, Arabic, Hebrew, Persian, Kabyle per lexicon, in el-ar-he-fa-kab.md
de, nl, fy, da, sv, nb (and no), nn German, Dutch, West Frisian, Danish, Swedish, Norwegian per lexicon, in germanic.md
fi, et, hu, tr, az Finnish, Estonian, Hungarian, Turkish, Azerbaijani per lexicon, in uralic-turkic.md
ru, uk, bg, pl, cs, sk, sl, hr Russian, Ukrainian, Bulgarian, Polish, Czech, Slovak, Slovene, Croatian per lexicon, in slavic.md
fr (fr-BE and fr-CH have their own lexicons), it, ro, eu, id French, Italian, Romanian, Basque, Indonesian per lexicon, in fr-it-ro-eu-id.md
es, gl, ca, ast, an, oc, mwl Spanish, Galician, Catalan, Asturian, Aragonese, Occitan, Mirandese per lexicon, in ibero.md

Each language page gives the default scale of each lexicon. Pass scale="long" or scale="short" to read the large scale words in the other scale. Another language tag raises UnsupportedLanguageError, a subclass of ValueError.

Each language is one JSON file in numtag/lexicon/. Each word and each setting in the file carries the source of its rule.

Documentation

Guide Content
docs/index.md The map of the documentation
docs/getting-started.md First calls and common tasks
docs/api.md Every public name
docs/values.md The value strings and how to convert them
docs/number-scale.md Long and short scale
docs/confidence-and-homographs.md Articles, words with two meanings, low scores
docs/rulings.md The reading rules, each with its source
docs/languages/ What each language reads
docs/adding-a-language.md The lexicon format and the gates for a new language
docs/design.md How the tagger works inside
docs/benchmarks.md Accuracy, memory and time against ovos-number-parser
docs/roadmap.md Planned features

License

Apache-2.0.

Metadata

Release files for numtag 0.10.10a1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for numtag 0.10.10a1
File Size Uploaded Provenance Actions
numtag-0.10.10a1.tar.gz 385.7 kB Without provenance Details for numtag-0.10.10a1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for numtag 0.10.10a1
File Interpreter ABI Platform Actions
numtag-0.10.10a1-py3-none-any.whl Python 3 none any Details for numtag-0.10.10a1-py3-none-any.whl

Total release size: 723.0 kB

Release files / numtag-0.10.10a1.tar.gz

Download URL numtag-0.10.10a1.tar.gz
Size 385.7 kB
Tags Source
SHA-256 checksum
How to use checksums
08335f21ed01cda01bdec1ec599a6cbf37b98b42d88d9444121d0bf5f644109e
BLAKE2b-256 checksum
How to use checksums
ed4efeff8745c4dc8125575fdbb533cfed2a21f1f317280d149b27889bbc951d
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.13.16

Release files / numtag-0.10.10a1-py3-none-any.whl

Download URL numtag-0.10.10a1-py3-none-any.whl
Size 337.3 kB
Tags Python 3
SHA-256 checksum
How to use checksums
14df14a96deb6ca960a2f7ad4be29c11186550014520c0d29cf0c4b71dde44c4
BLAKE2b-256 checksum
How to use checksums
c3a6f5aa6726f984ba3059bc9c29e718cdb2013ee4e762f97ed623ba1fdb8e89
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.13.16

Security for numtag 0.10.10a1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release
Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page