Functional Generative Description¶

A stratified dependency-based linguistic framework that links surface form to a tectogrammatical dependency representation integrating valency, semantic roles, coreference, and topic–focus articulation.

Version: v2 · 2026-09-06 · History
Domain-specific #: 1897
Origin domain: linguistics
Subdomain: functional dependency grammar
Aliases: FGD, Functional Generative Description framework

Core Idea¶

Functional Generative Description (FGD) is a formal linguistic framework developed in the Prague tradition. It represents an utterance through linked descriptive layers and treats the tectogrammatical layer as the underlying representation of sentence meaning. At that layer, lexical meaning units are nodes in a dependency tree, labeled with deep syntactic functions and enriched with valency, coreference, grammatical meaning, and information structure.^[1]

FGD is generative in the technical sense that its system of units and relations specifies how representations at one layer correspond to those at another. It is functional because communicative organization—especially topic–focus articulation—is part of the grammatical description rather than an optional discourse gloss.^[2]

The recognition invariant is linked stratified representations + dependency organization + tectogrammatical semantic structure + integrated topic–focus articulation.

Structural Signature¶

A declared sequence of linguistic layers rather than one undifferentiated parse.
Morphemic and surface-syntactic representations below a tectogrammatical representation.
Dependency trees whose nodes represent lexicalized meaning units.
Head–dependent edges carrying deep syntactic relations or functors.
Valency frames constraining the arguments licensed by lexical predicates.
Restoration or representation of semantically present elements omitted at the surface.
Coreference and grammatical-semantic attributes represented above surface syntax.
A horizontal ordering that encodes contextual boundness and topic–focus articulation.
Explicit mappings between adjacent layers.
An annotation interpretation sufficiently operational for treebank construction.
Separation between analytical surface syntax and tectogrammatical deep syntax.
A theory-to-corpus feedback loop in which annotated data tests and refines the representation.

What It Is Not¶

FGD is not a synonym for all dependency grammar. Dependency grammar supplies the head–dependent carrier, but FGD adds a named stratification and a particular deep representation. It is not merely a treebank annotation format: the Prague Dependency Treebank operationalizes the theory, while the theory supplies the distinctions that the annotation records.^[3]

It is also not equivalent to transformational-generative grammar despite the shared word generative. Its architecture, Prague functional commitments, dependency organization, and treatment of information structure are different.

Scope of Application¶

FGD supports theoretical syntax and semantics, valency lexicons, discourse and information-structure analysis, multilingual comparison, and richly annotated corpora. Its best-known operationalization is the Prague Dependency Treebank, where analytical and tectogrammatical annotations allow researchers and parsers to compare surface realization with deeper dependency structure.^[4]

The framework can guide computational analysis, but an implementation must declare which version of the annotation guidelines it follows because inventories and operational conventions evolve.

Clarity¶

An FGD analysis should state the layer, node inventory, dependency labels, valency source, treatment of ellipsis, coreference policy, and topic–focus ordering convention. A surface dependency tree must not be presented as if it were already tectogrammatical.

Manages Complexity¶

Layering prevents phonological, morphological, surface-syntactic, semantic, and discourse distinctions from being forced into one overloaded structure. Correspondence links preserve their relationship, while the tectogrammatical layer concentrates those features needed for meaning-oriented comparison across different surface realizations.

Abstract Reasoning¶

Segment and analyze the observed utterance at lower linguistic layers.
Construct the analytical dependency representation of surface syntax.
Identify lexical meaning units and restore licensed deletions where the theory requires them.
Project the structure into a tectogrammatical dependency tree.
Assign valency-based functors and grammatical-semantic attributes.
Resolve or mark coreference relations.
Encode contextual boundness and topic–focus articulation in deep word order.
Validate correspondence among layers and against the annotation manual.

Knowledge Transfer¶

The portable lesson is to separate representational levels while keeping explicit maps among them: surface evidence and semantic interpretation need not occupy the same structure. The proposed immediate parent is Dependency Grammar.

Examples¶

Active and passive clauses. Different surface configurations can be related to comparable tectogrammatical predicate–argument structures while voice remains represented as grammatical meaning.

Pro-drop or ellipsis. A semantically licensed participant absent from the surface may receive a tectogrammatical node under declared reconstruction rules.

Topic and focus. Two clauses with similar dependency content but different contextual organization can differ in tectogrammatical ordering and information-structure attributes.

Structural Tensions¶

Surface fidelity versus deep normalization.
Linguistic theory versus annotation reproducibility.
Rich semantic representation versus annotation cost.
Cross-linguistic portability versus Czech-developed categories.
Dependency hierarchy versus horizontal information structure.
Explicit reconstruction versus analyst over-interpretation.

Structural–Framed Character¶

Stratification, correspondence, and dependency are structural. Morphemes, valency, tectogrammatical functors, coreference, and topic–focus articulation are linguistic frame.

Structural Core vs. Domain Accent¶

The portable core is a multi-layer representation connected by declared transformations. The constitutive accent is FGD's particular Prague dependency architecture and its claim that information structure belongs inside the deep grammatical representation.

Dependency Grammar is the proposed immediate parent. Representation, Parsing, Interpretation, Hierarchy, Mapping, and Context are related primes or abstractions.

The prospective queue contains one strict edge to domain_specific:dependency_grammar. No live DAG mutation is authorized.

Relationships to Other Abstractions¶

Current abstraction Functional Generative Description Domain-specific

Parents (1) — more general patterns this builds on

Functional Generative Description is a kind of Dependency Grammar Domain-specific

Dependency Grammar is the proposed immediate parent.

Hierarchy path (1) — routes to 1 parentless root

Functional Generative Description → Dependency Grammar → Dependency

Neighborhood in Abstraction Space¶

Functional Generative Description sits in a sparse region of the domain-specific corpus (82^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Dependency grammar as a whole.
Phrase-structure or transformational-generative grammar.
The Prague Dependency Treebank as a corpus artifact.
Any generic deep structure.
Topic–focus articulation considered without the rest of FGD.
A surface dependency parse relabeled as tectogrammatical.

References¶

[1] Institute of Formal and Applied Linguistics, Charles University, “Functional Generative Description,” overview of the framework and its layer architecture, https://ufal.mff.cuni.cz/node/939. registry ↩

[2] Petr Sgall, Eva Hajičová, and Jarmila Panevová, The Meaning of the Sentence in Its Semantic and Pragmatic Aspects (D. Reidel, 1986). registry ↩

[3] Jan Hajič et al., Prague Dependency Treebank 2.0, Linguistic Data Consortium, 2006, describing the multilevel annotated corpus grounded in FGD. registry ↩

[4] Prague Dependency Treebank 2.0, “Basic Principles of Sentence Representation at the Tectogrammatical Level,” https://ufal.mff.cuni.cz/pdt2.0/doc/manuals/en/t-layer/html/ch02.html. registry ↩