FastText Language Detection

1.0.5 · maintenance · verified Wed Apr 15

fasttext-langdetect is a Python wrapper for Facebook's FastText language identification model. It offers fast and up to 95% accurate language detection across over 170 languages. The library, currently at version 1.0.5 (last released in January 2023), provides a straightforward interface for identifying the language of a given text string. It downloads the necessary FastText model on its first use.

Warnings

gotcha The FastText language model files are downloaded on the first call to the `detect()` function. This requires an active internet connection and sufficient disk space (the full model is ~126MB, the low-memory version is ~917KB). Subsequent calls will use the cached models.
Fix: Ensure network connectivity and appropriate disk space for the initial model download. The model typically caches in a temporary system directory.
gotcha The `low_memory` parameter (defaulting to `False`) controls which model is used. Setting `low_memory=True` loads a compressed model that uses less memory but may result in slightly lower detection accuracy compared to the full model (`low_memory=False`).
Fix: Choose `low_memory=True` for memory-constrained environments at the cost of slight accuracy, or `low_memory=False` (default) for maximum accuracy.
gotcha Language detection accuracy can be reduced for very short text inputs (e.g., single words, short phrases) or extremely long inputs. FastText models are generally optimized for text segments around 10-80 characters for optimal performance.
Fix: For short texts, results may be less reliable. Consider providing more context if possible. For very long texts, breaking them into smaller, meaningful segments might improve accuracy.
gotcha Pre-trained FastText models are often trained on clean, well-structured text. Noisy inputs (e.g., text with spelling errors, unusual capitalization, slang, or mixed languages/code-switching) can lead to reduced detection accuracy.
Fix: Consider pre-processing input text (e.g., lowercasing, basic cleaning) if dealing with informal or noisy user-generated content to potentially improve results.

Install

pip install fasttext-langdetect Install via pip

Imports

detect
```
from ftlangdetect import detect
```

Quickstart

This quickstart demonstrates how to import the `detect` function and use it to identify the language of a given text. It shows examples for both the full accuracy model (default) and the low-memory model.

from ftlangdetect import detect

# Detect language with default settings (low_memory=False for higher accuracy)
result_full = detect(text="Bugün hava çok güzel")
print(f"Full model result: {result_full}")

# Detect language with low_memory option (smaller model, slightly less accurate)
result_low_memory = detect(text="Bugün hava çok güzel", low_memory=True)
print(f"Low-memory model result: {result_low_memory}")

# Example with English text
english_text = "Hello, world! How are you?"
result_en = detect(text=english_text)
print(f"English text result: {result_en}")

view raw JSON →