lowresnltk 1.1.9

Low Resource NLTK

LowResNLTK

A low-resource Natural Language Processing toolkit.

Quick Inference Without Training

Pretrained model achieved an F1 Score of 97% in all classes.

POS Tagging

from lowresnltk import POSTagger

# Simple usage
tags = POSTagger.tag('আমি ভালো আছি')

Sentence Classification

from lowresnltk import SentenceClassifier

# Simple usage
label = SentenceClassifier.classify('আমি ভালো আছি')

Training Custom Models

Data Format Requirements

Column	Description	Example
Sentence	Full Bengali sentence	সন্ধ্যায় পাখিরা বাসায় ফেরে
Labels	Sentence type	Simple
POS	List of POS tags	['ক্রিয়া', 'বিশেষ্য', 'বিশেষ্য', 'অব্যয়']
Words	List of words	['সন্ধ্যায়', 'পাখিরা', 'বাসায়', 'ফেরে']

Example Dataset: https://huggingface.co/datasets/abkafi1234/POS-Sentence-Type

The code is Language Agnostic So Any Language will work. if the proper structure is followed

Train POS Tagger

import pandas as pd
from lowresnltk import POSTagger

# Load your data
data = pd.read_csv('Bangla.csv')

# Initialize and train
pt = POSTagger(data)
pt.train()

# Test prediction
result = pt.predict('আমি ভালো আছি')

Train Sentence Classifier

from lowresnltk import SentenceClassifier

# Load your data
data = pd.read_csv('Bangla.csv')

# Initialize and train
sc = SentenceClassifier(data=data)
sc.train()
result = sc.predict('আমি ভালো আছি')

Text Normalization Evaluator

A Unified Multi-Metric Framework for Evaluating Semantic Fidelity in Text Normalization*

Evaluate how stemming or lemmatization algorithms affect your text both structurally and semantically, beyond just compression ratios.

Python Example

from lowresnltk import NormalizationEvaluator
from sentence_transformers import SentenceTransformer
from sklearn.ensemble import RandomForestClassifier

# 1. Prepare your parallel datasets and labels
original_texts = ["The quick brown foxes are jumping.", "Beautifully painted canvases."]
normalized_texts = ["The quick brown fox be jump.", "Beauti paint canvas."]
labels = [0, 1]

# 2. Setup your evaluation models
classifiers = [RandomForestClassifier()]
embedding_model = SentenceTransformer("csebuetnlp/banglabert") 

# 3. Initialize and run the evaluator
evaluator = NormalizationEvaluator(
    texts_original=original_texts,
    texts_normalized=normalized_texts,
    labels=labels,
    classifiers=classifiers,
    embedding_model=embedding_model
)

# Returns CR, IRS, AES, ANLD, and MPD scores
results = evaluator.evaluate_all()
print(results)

Model Configuration

Default model paths:

• POS Tagger: ~/.lowresnltk/POSModel/
• Classifier: ~/.lowresnltk/ClassifierModel/

Installation

pip install lowresnltk

License

MIT License