Minimal recursion semantics¶
A flat, constraint-based computational-semantic representation using elementary predications, shared variables, handles, and scope constraints to preserve unresolved readings.
Core Idea¶
Minimal Recursion Semantics (MRS) is a flat, constraint-based representation of linguistic meaning for computational grammars. It records elementary predications, their shared variables, labels or handles, a distinguished top and index, and handle constraints while leaving compatible scope readings underspecified.
Instead of building one deeply nested formula during parsing, MRS collects local semantic contributions and restricts how their handles may outscope one another. Constraints such as qeq permit quantifier and operator scope to remain unresolved without accepting every logically possible arrangement. A later task can resolve scope when context or application demands it.
Operationally, Elementary predications encode minimal semantic relations and arguments. Variables and indices share individuals, events, and properties across predications. Handles and labels name predications and scopal positions. Top and index identify the representation’s entry handle and main semantic variable. Handle constraints restrict admissible scope without enumerating all readings. Grammar composition assembles the representation from lexical and phrasal contributions.
Scope of Application¶
MRS is used with computational grammar frameworks such as HPSG and in parsing, generation, machine translation, semantic transfer, and grammar engineering. Implementations can vary in predicate inventories and feature structures while preserving the EP-and-handle architecture.
MRS is not one fully scoped first-order formula, a bag of predicates with no argument links, a syntactic parse tree, or discourse representation theory generally. Flatness does not mean absence of structure: variables and handle constraints encode relations needed to reconstruct licensed readings.
Illustrative cases include: A quantified sentence is represented by EPs for predicates and quantifiers, shared individual variables, labeled handles, and qeq constraints admitting the intended scope alternatives. A grammar generator consumes an MRS, realizes a compatible sentence, and preserves semantic links across alternative surface forms.
Clarity¶
The abstraction separates compositional contribution from final scope resolution. It explains how a parser can preserve ambiguity compactly, how generation can share the same semantics, and why an MRS must be evaluated through the object-language readings it licenses.
Manages Complexity¶
Quantifier and modifier combinations can create many scoped formulas. MRS compresses this family into one flat representation plus constraints, avoiding premature enumeration. The economy shifts work to constraint validation and later resolution rather than eliminating ambiguity.
Abstract Reasoning¶
For each lexical or constructional contribution, create EPs with labeled handles and typed arguments. Share variables for common entities or events, set the top handle and main index, and add qeq or other constraints required by composition. Verify that intended scoped readings can be reconstructed and unintended ones are excluded. Resolve scope only when downstream evidence or task requirements justify a choice.
Knowledge Transfer¶
MRS transfers across grammars and languages when predicate inventories, types, and composition rules are adapted while handle semantics remain intact. A resolved output can transfer to theorem proving or execution, but the underspecified representation should not be mistaken for one reading. Flat graph-like storage elsewhere is only analogous unless it preserves the MRS semantic constraints.
Relationships to Other Abstractions¶
Current abstraction Minimal recursion semantics Domain-specific
Parents (1) — more general patterns this builds on
-
Minimal recursion semantics is a kind of Representation Prime
Minimal recursion semantics is a strict kind of Representation: A flat, constraint-based computational-semantic representation using elementary predications, shared variables, handles, and scope constraints to preserve unresolved readings.
Hierarchy path (1) — routes to 1 parentless root
- Minimal recursion semantics → Representation → Abstraction
Neighborhood in Abstraction Space¶
Minimal recursion semantics sits in a sparse region of the domain-specific corpus (89th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Logic form — 0.81
- Nota Accusativi — 0.80
- Predicate transfer — 0.80
- Functional Generative Description — 0.80
- Well-Formed Formula — 0.80
Computed from structural-signature embeddings · 2026-10-08