Trigram-Based Text Generation

Overview

This project implements a text generation model using a trigram-based approach. The model learns the probabilities of word sequences from input text and generates realistic sentences by predicting the next word based on the current context.

Features

• Trigram Model: Builds a model based on word triplets (trigrams).
• Probability-Based Generation: Generates text using probabilities calculated from trigram frequencies.
• Top 4 Candidate Selection: Randomly selects the next word from the top 4 most probable candidates to add diversity.
• Random Start: The model can begin generation from a random starting sequence in the training data.

Requirements

• Python 3.7+
• Libraries:
  • heapq (standard library)
  • random (standard library)
  • collections (standard library)
  • re (standard library)
• Dataset:
  • WikiText (for training large-scale models)

Installation

1. Clone the repository:

   git clone https://github.com/dhruv-pharasi/Trigram_TextGeneration.git

2. Ensure you have Python installed.
3. Install additional libraries if required (e.g., for preprocessing).

Usage

Training the Model

1. Provide a text corpus as a string.
2. Preprocess the text (clean, tokenize, etc.) and feed it into the model.
3. Build trigram and bigram frequency dictionaries:

   trigram_counts, bigram_counts = build_trigram_model(words)

Generate Text

• Live Text Generation with Random Start:

  starting_bigrams = list(trigram_counts.keys())
  
  start_seq = " ".join(random.choice(starting_bigrams))
  generate_text(trigram_counts, bigram_counts, start_seq, length=15)

• Customize Starting Sequence:

  start_seq = "once upon"
  generate_text(trigram_counts, bigram_counts, start_seq, length=15)

Example

Input Corpus

Once upon a time, there was a brave princess who fought dragons. The princess loved adventures.

Sample Output

once upon a time there was a brave princess who loved adventures and fought dragons

Files

• Trigram_TextGeneratorModel.ipynb: Contains the implementation of the trigram-based text generation.
• README.md: Documentation for the project.

Future Enhancements

• Add support for larger n-grams (e.g., 4-grams, 5-grams).
• Implement smoothing techniques (e.g., Laplace smoothing) for better handling of unseen sequences.
• Integrate with external APIs for training on large datasets.
• Provide a graphical interface for interactive text generation.

Author

Dhruv Pharasi