Bilingual-Sentence-Aligner

The code in this repo could be utilized to find matching sentence pairs between two TXT files, by making use of Google's recent LaBSE model. This is a simplified implementation of the same using PolyFuzz library.

To install necessary packages for the script, run:

pip install -r requirements.txt

To view script usage help from terminal, run:

python3 [scriptname].py -h

exec_alignfiles.py

The script accepts 2 TXT files as input and starts finding matching pairs using LaBSE. Along with TXT files, file encoding type and output filename must be specified.

To initiate the process:

python3 bilingual_sentence_aligner.py -i1 "input1.txt" -i2 "input2.txt" -e "utf-8" -o "out.csv"

NOTE: The Two text files are expected to have tokenized sentences for pairing.

The output will be a CSV file with 3 columns:

• COLUMN-1 : Sentence from TXT file 1
• COLUMN-2 : Matching Sentence from TXT file2 corresponding to COLUMN-1
• COLUMN-3 : Similarity between COLUMN-1 & COLUMN-2

faiss_aligner.ipynb

Notebook contains faiss based sentence aligner example. This approach could be utilized to align lakhs of sentences in minimal time.

Detailed article here