Abstract
This paper formalizes traditional Bangla grammar using the mathematical tools of graph theory. By mapping the structural logic of classical linguistics onto directed acyclic graphs (trees), the authors propose a model for language representation that sits between context-free and indexed grammars in terms of generative power. This approach bridges the gap between traditional South Asian grammatical descriptions and modern formalisms, providing a rigorous encoding of hierarchical and linear principles.
Introduction
Modern computational linguistics is largely dominated by Western frameworks such as Transformational-Generative (TG) grammar or Tree Adjoining Grammar (TAG). However, traditional South Asian grammatical traditions, particularly those described by scholars like Suniti Kumar Chatterji, offer a rich and alternative structural logic. This paper investigates how these traditional descriptions can be rigorously encoded using graph-theoretic concepts to understand how grammatical information is structurally "packed" into Bangla, from the sentence level down to internal word morphology.
Graph-Theoretic Sentence Representation
The researchers define the Bangla sentence as a directed acyclic graph—specifically a tree—where nodes represent grammatical categories and edges represent "embedding" relationships. This allows for a direct mapping of traditional Bangla concepts like uddeśya (subject), bidheya (predicate), samprasāraka (elaborator/adjunct), and pūraka (complement) onto formal tree structures.
A key innovation in this model is the concept of "maximality." To distinguish between multiple instances of the same category (e.g., different levels of a predicate), a node is considered "maximal" when its syntactic and semantic requirements, such as a transitive verb's need for a complement, are fully satisfied. This provides a mathematical way to define what constitutes a "complete" phrase.
Word-Internal Morphological Structures
Unlike many models that focus solely on sentence-level syntax, this graph-theoretic approach extends to the internal structure of words. Words are decomposed into prātipadika (core meaning), bibhakti (relational markers), prakriti (roots), and avayava (formatives).
As shown in the nominal constituent tree, the model captures the hierarchical relationship between the root and its various satellites and classifiers. This lexicalized grammar suggests that lexical items are not merely strings of characters but are associated with functional structural descriptions; the word itself contains the "blueprint" for the tree it can inhabit.
Theoretical Divergence and Generative Capacity
The proposed model highlights several interesting departures from standard Western linguistic theories. Most notably, it differs from X-Bar theory regarding the placement of "elaborators" (adjuncts). In this proposal, elaborators are kept outside the maximal projection, whereas X-Bar theory typically includes them within its scope.
Furthermore, the authors argue that only subtrees rooted in "maximally projected" nodes represent syntactically and semantically independent expressions. In terms of formal language theory, the model's "weak generative capacity" places it between context-free grammars and indexed grammars, making it a close relative to Tree Adjoining Grammar (TAG).
Conclusion
The authors successfully demonstrate that the traditional grammar of Suniti Kumar Chatterji can be translated into a modern graph-theoretic model without losing its descriptive nuances. While the current framework provides a robust foundation, future work is needed to integrate anusarga (postpositions) and resolve "finitely ambiguous" categories like certain classifiers. Ultimately, this work paves the way for a finer-grained analysis and potential computational implementation of traditional Bangla grammar.
Summary updated