N-gram Sentence Generator

A machine learning project that generates random sentences using n-gram language models trained on the Brown Corpus.

Overview

This project implements an n-gram language model to generate syntactically plausible sentences. It processes the Brown Corpus from NLTK, builds n-gram frequency models, and uses them to generate new sentences by selecting words based on probability distributions derived from the training data.

Features

• Flexible n-gram support: Generate sentences using bigrams (n=2), trigrams (n=3).
• Brown Corpus training: Leverages NLTK's Brown Corpus for robust linguistic patterns
• Customizable output: Control the number of sentences generated and maximum sentence length
• Statistical word selection: Uses n-gram frequency counts to select contextually appropriate next words

Requirements

• Python 3.x
• NLTK (Natural Language Toolkit)

Example

Enter the number of sentences to generate (M): 3
Enter the n-gram model to use (2 for bigram, 3 for trigram): 2
Enter the maximum length of a sentence (maxLen): 15

This will generate 3 sentences using a bigram model with a maximum of 15 words each.

How It Works

1. Preprocessing: The Brown Corpus is tokenized and converted to lowercase, with non-alphanumeric characters removed
2. N-gram Model Building: Frequency counts are collected for all n-grams in the corpus
3. Sentence Generation:
   • Start with a random n-gram from the training set
   • Iteratively select the next word based on which word most frequently follows the current context in the training data
   • Continue until the maximum length is reached or no valid continuation exists

Project Structure

NgramSentenceGenerator/
├── main.py          # Main script with all functions
└── README.md        # Project documentation

Functions

• preprocess_corpus(): Loads and preprocesses the Brown Corpus
• build_ngram_model(sentences, n): Creates frequency counts for all n-grams
• generate_sentence(ngram_model, vocabulary, n, max_len): Generates a single sentence
• main(M, N, maxLen): Orchestrates the sentence generation process

Limitations

• Generated sentences may not always be grammatically perfect or semantically meaningful
• Uses greedy word selection (always picks the most frequent next word) rather than probabilistic sampling
• Performance may vary depending on n-gram size and corpus coverage

Future Improvements

• Implement probabilistic sampling instead of greedy selection
• Add support for custom training corpora
• Cache n-gram models for faster repeated runs
• Add filtering for more natural-sounding output
• Implement smoothing techniques for better generalization