Recognizers-Text

1.0.2a2 verified Fri May 01 auth: no python

Recognizers-Text is a Microsoft-maintained NLP library for recognizing and normalizing numbers, units, date/time, phone numbers, emails, and URLs from unstructured text. It supports multiple languages (English, Spanish, French, Chinese, Japanese, and more) via separate per-culture packages. The Python package is at version 1.0.2a2 (alpha) and lags behind the .NET and JavaScript releases (v1.8.x). Releases are irregular.

pip install recognizers-text

Common errors

error ModuleNotFoundError: No module named 'recognizers_number' ↓

cause Attempting to import from sub-packages that do not exist as standalone modules.

fix

Use 'from recognizers_text import recognize_number' instead.

error AttributeError: type object 'Culture' has no attribute 'en-us' ↓

cause Passing a string like 'en-us' instead of the Culture class attribute.

fix

Use Culture.English or Culture.EnglishUs (if available). Check Culture.__dict__ for valid options.

error KeyError: 'values' ↓

cause Accessing result.resolution['values'] when resolution is None or not a dict (e.g., when no match or different type).

fix

Check that result.resolution is a dict and contains 'values' key. Use result.resolution.get('values', []) or handle empty cases.

error ImportError: cannot import name 'recognize_datetime' from 'recognizers_text' ↓

cause Older versions of the package may not have the function or the package is installed incorrectly.

fix

Upgrade to the latest version: pip install --upgrade recognizers-text

Warnings

gotcha The Python package (1.0.2a2) is an alpha release and significantly outdated compared to the .NET/JS versions (v1.8.x). Many languages and features (e.g., phone numbers, URLs) are missing. ↓

fix If you need the latest recognizers, use the .NET or JavaScript libraries instead.

gotcha All recognizer functions are in the same 'recognizers_text' top-level module. Do not import from sub-packages like 'recognizers_number' or 'recognizers_datetime' — they are not directly importable. ↓

fix Use 'from recognizers_text import recognize_number, recognize_datetime, ...'.

gotcha Culture is a class, not an enum or string. You must use Culture.English, Culture.Spanish, etc. Passing a string like 'en-us' will raise an AttributeError. ↓

fix Always use Culture attributes. Example: Culture.English.

deprecated The 'recognizers_text' package does not include the optional 'datatypes-timex-expression' dependency by default. Without it, accessing 'result.resolution' as a Timex object will fail. ↓

fix Install with extra: pip install recognizers-text[datatypes-timex-expression]

gotcha Recognition results may return empty lists for unrecognized text; they do not raise exceptions. Always check the result length. ↓

fix if not results: # handle no match

gotcha The 'recognizers_text' package name is slightly different from the GitHub repo name 'Recognizers-Text' (capitalized with hyphen). Import uses underscores. ↓

fix Use 'import recognizers_text' (all lowercase, underscore).

Install

pip install recognizers-text[datatypes-timex-expression]

Imports

Culture

wrong

from recognizers_text.culture import Culture

correct

from recognizers_text import Culture

Culture is a class within recognizers_text, not a submodule.

recognize_number

wrong

from recognizers_number import recognize_number

correct

from recognizers_text import recognize_number

All recognizers are in the 'recognizers_text' top-level package.

recognize_datetime

from recognizers_text import recognize_datetime

ModelResult

wrong

from recognizers_text.model import ModelResult

correct

from recognizers_text import ModelResult

ModelResult is exported from recognizers_text directly.

Quickstart

Recognizes a date from an English sentence and prints the recognition result.

from recognizers_text import Culture, recognize_datetime

# Recognize dates in English text
results = recognize_datetime("I'll be there on 2025-06-15", Culture.English)
for result in results:
    print(f"Text: {result.text}, Type: {result.type_name}, Resolution: {result.resolution}")

# Example output:
# Text: 2025-06-15, Type: datetimeV2.date, Resolution: {'values': [{'timex': '2025-06-15', 'type': 'date', 'value': '2025-06-15'}]}