cordex 1.0.0

Parser for collocability

Instructions

Necessary files:

• corpus in conllu or TEI format
• file with structures definitions
• Python 3.5+

We suggest usage of pypy3 for faster processing.

About

This library was developed to extract collocations from text in conllu or TEI format. Leveraging rules defined in the structure file (examples in tests/test_data/structures), it efficiently extracts and presents collocations.

Setup

You can run the library using either python3 or pypy3. For a better experience, we recommend utilizing virtual environments.

pip install -r requirements.txt

Running

Execution consists of three parts:

• setup
• execution
• collecting results

import cordex

# OPTIONAL: improve processing by downloading lookup lexicon (currently only available for Slovene) 
cordex.download('sl')
# setup
extractor = cordex.Pipeline("tests/test_data/structures/structures_UD.xml")
# execution
extraction = extractor("tests/test_data/input/ssj500k.small.conllu")
# collecting results
extraction.write("data/izhod.csv")

Setup

During this step you should provide processing settings. There is only one required parameter - path to structures file. Other parameters are optional.

Example

extractor = cordex.Pipeline("tests/test_data/structures/structures_UD.xml", statistics=False)

Parameters

structures

Required. Path to file with structures definitions. Examples of such file are available in tests/test_data/structures.

collocation_sentence_map_dest

Default value None. Optional parameter, that should contain path to file or folder (it has to be of the same type as path parameter in write() function). File or folder contain information about mappings between collocations and sentences. If this is set to None mappings will not be stored. When the path parameter in write() function leads to a file all mappings will also be stored in a file. Otherwise, program will create a directory at this destination and store mappings in multiple files inside this directory, one file per one syntactic structure. When you are using get_list function you may set this to True and it will return mappings in a second element of a tuple. If the file or folder on a given location already exist, they will be deleted before processing.

min_freq

Default value 0. Number that indicate how many occurrences in corpus a collocation needs to be present in results.

db

Default value None. Path to interprocessing sqlite database file (if there is no file in that location it will create a new file). It enables us to process corpus in steps, and stores half processed data. This parameter is useful for processing bigger corpora, as a failsafe system. Value None indicates that data will be stored only in memory.

overwrite_db

Default value False. This parameter should be used together with parameter db. When True it will overwrite old database file and start processing from the beginning.

jos_msd_lang

Default value en. Set this to sl when xpos tags are in Slovenian.

ignore_punctuations

Default value False. When this is True, results containing punctuations will not be shown.

fixed_restriction_order

Default value False. When this is True, results, where structure components are not in the same order as in structure definition files, will be ignored.

lookup_lexicon

Default value None. Path to lookup lexicon. Lexicon is used to improve representations when JOS system is used. Value None indicates that we are not using lookup lexicon.

lookup_api

Default value False. When this is True, program will use api to improve representations when JOS system is used. Lookup lexicon will be ignored in this case.

statistics

Default value True. Parameter that indicates whether we want statistics in output file or not.

lang

Default value sl. Parameter that enables postprocessing for specific languages. Should contain lowercased 2-letter country abbreviation.

jos_depparse_lang

Default value sl. When using JOS system, extraction will work with Slovenian (sl) or English (en) dependency parsing tags. This is not connected to UD dependency parsing in any way.

Execution

During this step extraction executes.

Example

extraction = extractor("tests/test_data/input/ssj500k.small.conllu")

Parameters

corpus

Required. Path to corpus we want to be processed. Data may be in conllu or TEI format. This path should point to either concrete file containing corpus or directory containing multiple files in the same format. Examples of such files/folders are available in tests/test_data/input.

Collecting results

We support two ways of obtaining results. You may use function write, to write results directly. You may also use function get_list, which will return a list with results.

Examples

Write example:

extraction.write("data/izhod.csv")

Get list example:

results = extraction.get_list(sort_by=1, sort_reversed=True)

Parameters

path

Required in write function. Destination where results are going to be written. If this is a path to a folder, results are going to be stored in multiple files, otherwise in a single file.

separator

Default value \t. Optional parameter in write and get_list functions, that tells us what output files should be separated by. For .csv files shis should be ,.

decimal_separator

Default value .. Optional parameter in write and get_list functions, that tells us what should be a separator for decimal numbers.

sort_by

Default value -1. Parameter in write and get_list functions indicating by which column processing results will be sorted. Value -1 indicates that results will be sorted by the order of structures given in structure definitions file.

sort_reversed

Default value False. This parameter is related to sort_by. When set to True, results will be sorted in reverse order of selected column.

Optional: download lookup lexicon

You may download lookup lexicon, that enhances representations of collocations. By default, the lexicon will be downloaded to ~/cordex_resources directory. You may change this to other location by specifying dir parameter in download function. Keep in mind that if custom path is selected, lexicon will only be used if an appropriate lookup_lexicon path is given during Setup step. If lexicon is on default path it will be used automatically. Lookup lexicon is currently only available for Slovenian language. You may download it using the following command. You may manually download lexicon from clarin repository.

cordex.download('sl')

Instructions for running on big files (ie. Gigafida)

You should run script using pypy3.

Suggested running with saved mysql file in tmpfs. Instructions:

sudo mkdir /mnt/tmp
sudo mount -t tmpfs tmpfs /mnt/tmp

If running on big corpuses (ie. Gigafida have database in RAM):

sudo mkdir /mnt/tmp
sudo mount -t tmpfs tmpfs /mnt/tmp
sudo mount -o remount,size=110G,noexec,nosuid,nodev,noatime /mnt/tmp