Python tool for processing and tagging sentences with IG 2.0 syntax with additional tools for text cleaning, preprocessing and postprocessing.
- Create a virtual environment:
python -m venv .env
- Activate the virtual environment:
source .env/bin/activate
- Install package
python -m pip install --upgrade pip
python -m pip install ig-tagger
ig-cli
Possible tasks are executed as shell commands on files:
ig-cli <task_type> <input_file_path> -o output_file_path --some-additional-option
To show information about possible commands, arguments and options execute:
ig-cli -h
Input:
Plain .txt file with text.
Output:
Plain .txt file with sentences separated by new empty lines or .tsv file with ['sentence no.', 'sentence_type', 'text'] columns
(with optional parameter --format=tsv)
Command:
ig-cli atomize input_text.txt
ig-cli atomize input_text.txt -o sentences.txt --split_type ml
ig-cli atomize input_text.txt --format txt
Optional parameters
- --format (txt/tsv)
- --output_file_path -o
- --split_type (ml/rule_based)
About:
Complex sentences with enumerations are splitted into atomic sentences when it is possible. (xxx xxx (a) ccc, (b) vvv” -> “xxx xxx ccc”, “xxx xxx vvv”).
Split type possible values: ‘ml’, ‘rule_based’. ML variant uses a special tool (Spacy library) for recognizing the beginnings and ends of sentences in text. Rule-based variant uses simple matching based on capital letter and period at the end of the sentence (regular expressions).
These two are different approaches and can give different results. The basic option is rule_based, ml can do better with lower quality text because of considering whole sentence structure (not only dots and capital letters).
Both splits recognize enumeration based on a, b, c… or 1, 2, 3… to split bigger sentences into smaller ones. Which is implemented as matching such expressions (xxx xxx (a) ccc, (b) vvv”) in the sentence, then splitting and constructing new sentences from extracted parts (“xxx xxx ccc”, “xxx xxx vvv”).
For example:
-
The employee is subject to (1) a Federal quarantine order related to COVID-19 (2) a Federal isolation order related to COVID-19.
-
The employee is subject to a Federal quarantine order related to COVID-19.
-
The employee is subject to a Federal isolation order related to COVID-19.
Sentences 2-3 are extracted from sentence 1 based on (1) (2) pattern.
Input:
Plain .txt file with sentences separated by two new lines (\n\n) or .tsv file with 3 columns ['sentence no.', 'sentence_type', 'text']. (Based on file extension)
Output:
.tsv file with 3 columns: ['sentence no.', 'sentence_type', 'text'].
Command:
ig-cli classify sentences.tsv
Optional parameters
- --output_file_path -o
About:
Sentences are classified as regulative (r) or constitutive (c). For this purpose, simple ML model is prepared trained on a small annotated dataset. The output file should be reviewed and corrected manually.
Input:
.tsv file with 3 columns ['sentence no.', text, 'sentence_type'] compatible with results of classify command.
Output:
.tsv file with tagged sentences
Command:
ig-cli tag classified_sentences.tsv -o tagged_sentences.tsv
Optional parameters
- --output_file_path -o
About:
Tagging is based on natural language processing with linguistic features recognition and rules constructed for mapping linguistic features to Institutional Grammar tags. Every sentence is analysed accordingly then results are saved with tags corresponding to each word token.
Input:
Output:
Command:
About:
Comparison between files (e.g. for quality/error assessment) is possible via other tools such as (diff - command line tool (use diff -h for detailed instruction), diffchecker - web tool)
- Sentence type classification - The ML model can be changed/retrained as a new file with serialized Python object with
.predict(self, sentences: List[str]) -> List[bool]method and returns True for regulative sentences. Corrected files can be gathered for building better classifier.
The package can be used within import igtagger with object-oriented operations included in igtagger.backend and file operations included in igtagger.frontend.
from igtagger import backend
backend.get_annotated_sentences(df)
Detailed documentation is in the PDF.
Project leaders: Anna Wróblewska, Bartosz Pieliński
The tool is based on the results of previous work on Institutional Grammar annotation: