SECOS

This repo is a modular python implementation of the SECOS algorithm for decomposing composite nouns.

Based on the SECOS algorithm:

original python implementation

original paper

However, the training data of the models have been distilled slightly to reduce the size of the models. More information on this can be found in the pretrained-models directory in the github repo. Typically though it involves trimming out words with low frequency counts, words with non-unicode characters etc.

Installation

From Github

pip install git+https://github.com/mhaugestad/noun-splitter.git -U

From Source

git clone
cd noun-splitter
pip install -e . -U

From Pip

pip install noun-splitter

Installing models:

The module relies on pretrained models to be passed in. These can be downloaded from command line as follows:

python -m secos download --model de

The command line tool also takes an optional argument --overwrite. This is to be used if you would like to redownload a model for whatever reason, as follows:

python -m secos download --model no --overwrite

Alternatively, you can download models directly from a python script or notebook like this:

from secos import Decomposition

Decomposition.download_model('de')

Available models and their names are:

Language	Model
Danish	da
German	de
English	en
Spanish	es
Estonian	et
Finnish	fi
Hungarian	hu
Latin	la
Latvian	lv
Netherland	nl
Norway	no
Swedish	sv

Basic Usage

from secos import Decomposition

model = Decomposition.load_model('de')

secos = Decomposition(model)

secos.decompose("Bundesfinanzministerium")

['bundes', 'finanz', 'ministerium']

Evaluation

The evaluation folder in the github repo includes code for the evaluation of the pretrained models.