Copying in Natural Languages, Context-Freeness, and Queue Grammars

Alexis Manaster-RamerACL 1986 · DOI

In his 1986 paper, Alexis Manaster-Ramer presents a foundational critique of the application of formal language theory to human speech. The central thesis is that the standard "Chomsky Hierarchy"—the traditional map for categorizing the complexity of languages—is fundamentally misaligned with the structural realities of natural language. While Context-Free Grammars (CFGs) have long been the industry standard for modeling syntax, Manaster-Ramer argues they are built on the wrong computational foundation: stacks (Last-In, First-Out), whereas human language fundamentally relies on queues (First-In, First-Out).

The Context-Free Debate and the "Finite" Trap

For decades, linguists have debated whether natural languages are strictly "Context-Free." A Context-Free Grammar is highly effective at handling nested structures, such as the English sentence: "The cat [that the dog chased] ran away." These structures are easily modeled using a stack, where the most recently opened dependency is the first one to be resolved.

However, CFGs struggle with two specific phenomena: "copying" (reduplication, or strings of the form $ww$) and "cross-serial dependencies" (found in languages like Dutch and Swiss German). Manaster-Ramer addresses two common counter-arguments used to defend the CFG model:

  1. The Finite Argument: Some argue that because human sentences are finite and vocabularies are limited, any natural language is technically "regular" and thus context-free. Manaster-Ramer dismisses this as a mathematical technicality that ignores the productive nature of language. Humans can invent a new word and immediately know how to apply a "copying" rule to it, suggesting the underlying grammar is not constrained by a fixed, finite list.
  2. The Symmetry Problem: CFGs are mathematically predisposed to generate "mirror-image" languages ($ww^R$). Curiously, human languages almost never use mirror-image structures. Conversely, CFGs are poor at generating "copying" languages ($ww$), which human languages use constantly. This asymmetry suggests that the Chomsky Hierarchy is the wrong "formal space" for linguistics.

Proposing Context-Free Queue Grammars (CFQG)

To better align formal models with linguistic reality, the paper proposes a new class of grammars: Context-free Queue Grammars (CFQG). Unlike a standard CFG where symbols are replaced in place or pushed onto a stack, a CFQG allows a rule to send a symbol to the end of a "queue" (the end of the string being rewritten).

This shift from LIFO to FIFO (First-In, First-Out) logic allows the grammar to naturally handle copying while avoiding the mirror-image structures that are absent in natural speech. Manaster-Ramer also shifts the focus from "weak generative capacity" (the final set of strings produced) to "classificatory capacity"—the ability of a grammar to represent the actual syntactic patterns and dependencies observed by linguists.

Cross-Linguistic Evidence for Reduplication

Manaster-Ramer demonstrates that copying is not a marginal quirk of a few "exotic" languages, but a pervasive and fundamental process. The paper draws on a wide variety of linguistic data:

  • English: Productive constructions such as "Linguistics shminguistics," "A dog is a dog is a dog," and "X or no X" patterns.
  • Bambara: Complex noun reduplication.
  • Swiss German & Dutch: Cross-serial verb constructions that follow a FIFO dependency pattern.
  • Other Languages: Tamil (kiduplication), Turkish (meduplication), Polish, and Sanskrit.

These examples illustrate that reduplication is used for emphasis, derision, quantification, and core syntax across unrelated language families.

Key Insights and Theoretical Impact

The "aha moment" of the paper is the realization that human linguistic processes prefer queues over stacks. We tend to process dependencies in the order they appear rather than nesting them deeply like a computer stack. By formalizing CFQGs, Manaster-Ramer demonstrates that it is possible to create a model that generates copying languages ($ww$) but—conjecturally—cannot generate mirror-image languages ($ww^R$). This aligns the formal mathematical model with actual human linguistic constraints.

Limitations and Future Work

While the paper successfully argues for a paradigm shift, it acknowledges several remaining challenges. The author admits that the current definition of CFQGs might be "profligate," potentially generating more structures than necessary or being computationally inefficient. Furthermore, while the paper pushes for queues, it acknowledges that "center-embedding" (which requires a stack) does exist in language. A truly adequate model would likely need to be a hybrid, explaining how stack-based and queue-based resources coexist in the human mind. Finally, the paper suggests that future work is needed to generalize these grammars to handle multi-copying (e.g., "Grammar is grammar is grammar").

Summary updated

BibTeX
@inproceedings{manasterramer1986copying,
  author = {Alexis Manaster-Ramer},
  title = {{Copying in Natural Languages, Context-Freeness, and Queue Grammars}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {1986},
  doi = {10.3115/981131.981145},
}
Figure