Grammar Formalisms
The richer grammar systems used to describe natural-language syntax beyond plain context-free rules — tree-adjoining, categorial, and unification-based grammars — that balance expressiveness against parseability.
Definition
A grammar formalism is a system for specifying the legal structures of a language, characterized by its generative power, its representational devices, and the complexity of parsing it.
Scope
Covers formalisms designed to capture syntactic phenomena that context-free grammars handle awkwardly: tree-adjoining grammar (TAG), combinatory categorial grammar (CCG), and unification- or feature-based grammars such as HPSG and LFG. It addresses the notion of mild context-sensitivity and the trade-off between linguistic adequacy and computational tractability. Parsing algorithms specific to constituency and dependency are covered separately.
Core questions
- What syntactic phenomena motivate going beyond context-free grammars?
- What is mild context-sensitivity and why is it considered the right power for language?
- How do unification and features let grammars share information across a structure?
- How do these formalisms trade expressiveness against efficient parsing?
Key concepts
- tree-adjoining grammar
- combinatory categorial grammar
- unification grammar
- feature structure
- mild context-sensitivity
- HPSG
- LFG
- subcategorization
Key theories
- Mild context-sensitivity
- A class of grammars, including TAG and CCG, that exceeds context-free power just enough to capture cross-serial dependencies while remaining polynomial-time parseable.
- Combinatory categorial grammar
- A lexicalized formalism in which words carry functional categories combined by a small set of combinators, tightly coupling syntax with compositional semantics.
- Unification-based grammar
- Formalisms such as HPSG that represent linguistic objects as typed feature structures combined by unification, capturing agreement and subcategorization declaratively.
History
Through the 1980s and 1990s, linguists developed formalisms richer than context-free grammars to capture agreement, long-distance dependencies, and the syntax–semantics interface. Joshi's tree-adjoining grammar formalized mild context-sensitivity, while CCG and HPSG offered lexicalized and unification-based alternatives that remain influential in both theoretical and computational work.
Debates
- Expressiveness versus tractability
- More powerful formalisms can describe more phenomena but risk intractable parsing; the field generally favors the least power sufficient to cover the data, hence interest in mildly context-sensitive grammars.
Key figures
- Aravind Joshi
- Mark Steedman
- Carl Pollard
- Ivan Sag
Related topics
Seminal works
- joshi1997
- steedman2000
- pollard1994
Frequently asked questions
- Why not just use context-free grammars for everything?
- Some constructions, like cross-serial dependencies in Dutch and Swiss German, provably cannot be generated by context-free grammars. Mildly context-sensitive formalisms add just enough power to handle them while staying efficiently parseable.