sparv-freeling 4.0

FreeLing plug-in for Sparv (Språkbanken's corpus annotation pipeline)

sparv-freeling

This is a plugin for the Sparv pipeline containing a wrapper for FreeLing. Please observe that this plugin has a more restrictive license than the Sparv piepeline!

This plugin allows you to run the Sparv pipeline and get sentence segmentation, tokenisation, baseform analysis, and part-of-speech annotations for the following languages:

• Asturian
• Catalan
• English
• French
• Galician
• German
• Italian
• Norwegian
• Portuguese
• Russian
• Slovenian
• Spanish

Furthermore Sparv will convert the FreeLing POS-tags into Universal POS tags and output them as a separate annotation.

Some of these languages (Catalan, English, German, Portuguese and Spanish) also support named-entity recognition.

Prerequisites

• Sparv pipeline
• FreeLing 4.2 and freeling-langs-4.2
• Python 3.6 or newer

Installation

Option 1: Installation from pypi with pipx:

pipx inject sparv-pipeline sparv-freeling

Option 2: Installation from GitHub with pipx:

pipx inject sparv-pipeline https://github.com/spraakbanken/sparv-freeling/archive/latest.tar.gz

Option 3: Manual download of plugin and installation in your sparv-pipeline virtual environment:

source [path to sparv-pipeline virtual environment]/bin/activate
pip install [path to the downloaded sparv-freeling directory]

Usage

The Sparv pipeline needs a config file describing your corpus and the desired output format. Please refer to the Sparv pipeline user manual for more details on config files and running Sparv.

Example input:

<text title="Example">
  This is an example for how to run Sparv.
</text>

Example command for creating xml with annotations:

sparv run

Result file:

<?xml version="1.0" encoding="UTF-8"?>
<text lix="20.00" title="Example">
  <sentence>
    <token baseform="this" pos="DT" upos="DET">This</token>
    <token baseform="be" pos="VBZ" upos="VERB">is</token>
    <token baseform="a" pos="DT" upos="DET">an</token>
    <token baseform="example" pos="NN" upos="NOUN">example</token>
    <token baseform="for" pos="IN" upos="ADP">for</token>
    <token baseform="how" pos="WRB" upos="ADV">how</token>
    <token baseform="to" pos="TO" upos="PART">to</token>
    <token baseform="run" pos="VB" upos="VERB">run</token>
    <token baseform="sparv" ne_type="person" pos="NP00SP0" upos="PROPN">Sparv</token>
    <token baseform="." pos="Fp" upos="PUNCT">.</token>
  </sentence>
</text>

Additional Info about Annotations

A full list of what analyses are supported for what languages can be found here:

https://freeling-user-manual.readthedocs.io/en/latest/basics/#supported-languages

Integrating dependency parsing

FreeLing supports dependency parsing for some languages. The output format is a bit cumbersome though.

Input:

This is a sentence.

Output:

DT/top/(This this DT -) [
  vb-be/modnorule/(is be VBZ -)
  sn-chunk/modnorule/(sentence sentence NN -) [
    DT/det/(a a DT -)
  ]
  st-brk/modnorule/(. . Fp -)
]

It is possible to write a new parser to handle this format but so far this has not been a priority for us.