alphabetic_sort

Sort numbers by the alphabetical order of their name in any language.

What is this?

Numbers, when written out as words, have a perfectly valid alphabetical order. Eight comes before five, which comes before four, which comes before nine, and so on. This order has nothing to do with numerical value — it is purely determined by how the words are spelled. Most of the time this is completely useless. Occasionally, it is exactly what you need.

Naturally, this is entirely language-dependent. In English, eight sorts first among 1–9. In German it is acht, which also sorts first (German and English agree on this one — enjoy it, it is rare). In French it is deux. Same numbers, completely different order. The numbers do not care. They never did.

This library takes a list of numbers and sorts them the way a dictionary would: by converting each number to its written word form in a given language, sorting those words alphabetically, and returning the original numbers in that new order.

from alphabetic_sort import alphabetic_sort

alphabetic_sort([1, 5, 12], lang="en_UK")
# → [5, 1, 12]
# because "five" < "one" < "twelve"

The result looks wrong at first glance. That is the point.

Origin

The idea came from a real-world encounter. A biologist colleague once sent over an Excel sheet containing a long list of sample numbers. They had sorted the column themselves — and because Excel, left to its own devices, had treated the values as text, the numbers ended up in alphabetical order: 1, 10, 100, 11, 12, 2, 20, 21... The biologist saw nothing wrong with this. The data was sorted, after all. Alphabetically sorted numbers are a thing that exists in the wild.

That conversation stuck. This library is the result.

A note on how this was built

This entire project — library, tests, CI pipeline, web UI, and documentation — was written by Claude on a Sunday, entirely from a cell phone, while going about a perfectly ordinary day. No laptop, no desk, no dedicated coding session. Just a series of prompts tapped out between whatever else the day had going on. The code is real, the tests pass, and the whole thing is published to GitHub.

Make of that what you will.

Yes, this is a joke. It works for real.

from alphabetic_sort import alphabetic_sort

alphabetic_sort([1, 5, 12], lang="en_UK")
# → [5, 1, 12]
# because "five" < "one" < "twelve"

Installation

pip install alphabetic_sort

Requires Python 3.11+.

Usage

from alphabetic_sort import alphabetic_sort

# Integers
alphabetic_sort([1, 5, 12], lang="en")       # → [5, 1, 12]

# Floats
alphabetic_sort([1.5, 0.5, 2.5], lang="en")  # → [1.5, 2.5, 0.5]
                                              # "one.." < "two.." < "zero.."

# French Belgian (septante/nonante, not soixante-dix/quatre-vingt-dix)
alphabetic_sort([11, 90], lang="fr_BE")       # → [90, 11]
# "nonante" < "onze"  (n < o)

alphabetic_sort([11, 90], lang="fr")          # → [11, 90]
# "onze" < "quatre-vingt-dix"  (o < q)

# German
alphabetic_sort(list(range(1, 13)), lang="de")
# → [8, 3, 1, 11, 5, 9, 6, 7, 4, 10, 2, 12]
# acht, drei, eins, elf, fünf, neun, sechs, sieben, vier, zehn, zwei, zwölf

Negative numbers

Negatives always sort before non-negatives. Among negatives, sorting is in reverse alphabetical order of the absolute value word ("minus" is ignored).

alphabetic_sort([-5, -1, -12, 3, 7], lang="en")
# → [-12, -1, -5, 7, 3]
#
# Negatives (rev alpha of abs): twelve > one > five  →  -12, -1, -5
# Positives (alpha):            seven < three        →   7, 3

Zero is treated as non-negative.

Intermediate steps

Pass return_words=True to get the words used at each stage:

sorted_nums, original_words, sorted_words = alphabetic_sort(
    [1, 5, 12], lang="en", return_words=True
)
# sorted_nums    → [5, 1, 12]
# original_words → ["one", "five", "twelve"]  (matches input order)
# sorted_words   → ["five", "one", "twelve"]  (matches output order)

Language codes

Pass any BCP 47 language tag — it is normalized automatically:

alphabetic_sort([1, 5, 12], lang="en_UK")   # same as "en"
alphabetic_sort([1, 5, 12], lang="en-GB")   # same as "en"
alphabetic_sort([1, 5, 12], lang="fr_BE")   # Belgian French (septante/nonante)
alphabetic_sort([1, 5, 12], lang="de_AT")   # same as "de"

Regional variants with genuinely distinct words are preserved:

Supported languages

56 locales via num2words:

am ar az be bn ca ce cs cy da de en en_IN en_NG eo es es_CO es_CR es_GT es_NI es_VE fa fi fr fr_BE fr_CH fr_DZ he hu id is it ja kn ko kz lt lv nl no pl pt pt_BR ro ru sk sl sr sv te tet tg th tr uk vi

from alphabetic_sort import get_supported_languages
get_supported_languages()  # → sorted list of all 56 locale keys

Performance and multithreading

The number-to-word conversion step (num2words) dominates runtime for large inputs. The library eliminates redundant calls by building a word cache before sorting: each unique number (and the absolute value of each negative) is converted exactly once.

The workers parameter exposes an optional ThreadPoolExecutor for that conversion step. The default is workers=1 (single-threaded).

On standard Python builds (GIL enabled — 3.11, 3.12, 3.13): threading adds overhead rather than removing it. The GIL serializes the num2words calls regardless of how many threads you spawn, while thread management still costs time. Leave workers=1 on these builds.

On free-threaded Python builds (3.13t, 3.14t+): the GIL is disabled and threads genuinely run in parallel. workers=4 (or higher) gives near-linear speedup on large inputs. This is the intended use case for the parameter.

# Free-threaded Python only — no benefit on standard builds
alphabetic_sort(large_list, lang="en", workers=4)

The CI benchmark workflow (benchmark.yml) measures this automatically across Python 3.11–3.15 with and without free-threading, and publishes results to the GitHub Actions step summary so the numbers are always up to date.

API

alphabetic_sort(numbers, lang, *, locale_aware=False, return_words=False, workers=1)

Raises: UnsupportedLanguageError, NumberConversionError, TypeError

get_supported_languages() → list[str]

Returns a sorted list of all supported locale keys.

Web UI

A demo web app is included in webui/:

pip install -r webui/requirements.txt
uvicorn webui.app:app --reload
# → http://127.0.0.1:8000

License

Apache 2.0 — see LICENSE.