MorphGen: Morphological Generators for Indian Languages

Generating inflectional wordforms and morphological features for 10+ Indian languages, facilitating synthetic dataset creation and system development.

Supported by: Oxford Languages

Core Objectives

1

Inflectional Generation

Generate all inflectional wordforms from a given lemma or wordform.

2

Morphological Features

Produce complete morphological features for each generated form.

3

Rule Application

Optionally output the specific rule applied during form generation.

Rule-Based System Architecture

In-House Rule Language

Built on regular expressions, the custom rule-expression language simplifies rule editing.

Language Support

Currently supports more than 10 Indian languages, with ongoing expansion.

Applications of Morphological Generators

Synthetic Dataset Generation

Create large-scale synthetic datasets for training NLP models.

Generative System Building

Develop generative systems adaptable to various hardware configurations.

Supported Indian Languages

Hindi

A widely spoken language in North India.

Bengali

Dominant in West Bengal and Bangladesh.

Assamese

Spoken in the Assam region.

Marathi

Predominant in Maharashtra.

Odia

Spoken in Odisha.

Nepali

Official language of Nepal and spoken in Sikkim.

Sanskrit

An ancient and classical language.

Punjabi

Spoken in the Punjab region.

MorphGen Language

The MorphGen language is a custom language designed on top of regular expressions for describing morphological rules in Indian languages. It provides a concise and intuitive way to specify complex patterns and transformations, leveraging the power of regular expressions while adding higher-level abstractions tailored for morphological analysis and suited for non-technical experts.

Synthetic Data Generation Process

1

Define Lemmas

2

Apply Rules

3

Generate Forms

4

Output Dataset

Advantages of Our Approach

1

Accuracy

High precision in morphological generation.

2

Flexibility

Easy to adapt to new languages and rules.

3

Efficiency

Fast generation of morphological forms.

4

Transparency

Rule-based System for synthetic data generation

© 2022-26 UnReaL-TecE LLP. All rights reserved.