** See https://github.com/bsolomon1124/pycld3 for a more update Python binding version. (And easier pip install) **

Compact Language Detector v3 (CLD3) Python Edition

• Model
• Installation
• Contact
• Credits

Included are Python bindings (via Cython) for cld3, a language classifying library used in Google Chrome. Building the extension does not require the Chromium repository, instead only these libraries need to be installed:

• Cython
• msgpack
• Protobuf (with headers and protoc)

To install the extension, from this repo, run pip install https://github.com/ThothMedia/cld3/archive/master.zip

Model

CLD3 is a neural network model for language identification. This package contains the inference code and a trained model. The inference code extracts character ngrams from the input text and computes the fraction of times each of them appears. For example, as shown in the figure below, if the input text is "banana", then one of the extracted trigrams is "ana" and the corresponding fraction is 2/4. The ngrams are hashed down to an id within a small range, and each id is represented by a dense embedding vector estimated during training.

The model averages the embeddings corresponding to each ngram type according to the fractions, and the averaged embeddings are concatenated to produce the embedding layer. The remaining components of the network are a hidden (Rectified linear) layer and a softmax layer.

To get a language prediction for the input text, we simply perform a forward pass through the network.

Installation

Building the Python wheel requires the protobuf compiler and its headers to be installed. If you run into issues with protobufs not compiling, just go into the src directory and run

mkdir -p cld_3/protos
protoc --cpp_out=cld_3/protos *.proto

To generate a python wheel (from the root of this repo):

python setup.py bdist_wheel

Builds have been tested with GCC9.0 on Ubuntu 18.04 and Apple Clang 11.0.0 on OSX 10.15 (Catalina Beta)

Python Usage

Here's some examples:

cld3.get_language("This is a test")
# LanguagePrediction(language='en', probability=0.9999980926513672, is_reliable=True, proportion=1.0)

cld3.get_frequent_languages("This piece of text is in English. Този текст е на Български.", 5)
# [LanguagePrediction(language='bg', probability=0.9173890948295593, is_reliable=True, proportion=0.5853658318519592), LanguagePrediction(language='en', probability=0.9999790191650391, is_reliable=True, proportion=0.4146341383457184)]

In short:

• get_language returns the most likely language as the named tuple LanguagePrediction. Proportion is always 1.0 when called in this way.
• get_frequent_languages will return the top number of guesses, up to a maximum specified (in the example, 5). The maximum is mandatory. Proportion will be set to the proportion of bytes found to be the target language in the list.

In the normal cld3 library, "und" may be returned as a language for unknown languages (with no other stats given). This library filters that result out as extraneous; if the language couldn't be detected, nothing will be returned. This also means, as a consequence, get_frequent_languages may return fewer results than what you asked for, or none at all.

Credits

Original authors of the code in this package include (in alphabetical order):

• Alex Salcianu
• Andy Golding
• Anton Bakalov
• Chris Alberti
• Daniel Andor
• David Weiss
• Emily Pitler
• Greg Coppola
• Jason Riesa
• Kuzman Ganchev
• Michael Ringgaard
• Nan Hua
• Ryan McDonald
• Slav Petrov
• Stefan Istrate
• Terry Koo

and Elizabeth Myers for the original Python bindings