Inferensys

Glossary

Predicate-Argument Structure

A linguistic framework representing a sentence as a predicate (typically a verb) and its associated arguments (such as agent, patient, and instrument) that define the core semantic relationships.
Developer reviewing multi-agent chat interface on laptop, agent conversation logs visible, casual coding session at WeWork desk.
SEMANTIC FRAMEWORK

What is Predicate-Argument Structure?

The linguistic framework representing a sentence as a predicate (typically a verb) and its associated arguments (such as agent, patient, and instrument) that define the core semantic relationships.

Predicate-argument structure is the formal linguistic representation of a sentence's core meaning, decomposing it into a predicate—the relational element, typically a verb—and its arguments, the noun phrases that complete the predicate's meaning. This framework answers 'who did what to whom' by identifying the semantic roles each participant plays in the event or state described by the predicate.

In computational linguistics, this structure is the target of Semantic Role Labeling (SRL), where systems map syntactic constituents to thematic roles like Agent, Patient, and Instrument. Resources such as PropBank and FrameNet provide annotated corpora that define verb-specific framesets, enabling models to disambiguate predicate senses and classify arguments with precision.

LINGUISTIC FRAMEWORK

Key Characteristics of Predicate-Argument Structures

The predicate-argument structure is the core semantic architecture of a clause, representing the central organizing relationship between a predicate and its participants. It abstracts away from surface syntax to capture who did what to whom.

01

Predicate as Semantic Anchor

The predicate—typically a verb, but also a predicative adjective or noun—serves as the relational hub of the clause. It determines the number, type, and semantic roles of its arguments through its valency or arity.

  • An intransitive verb like sleep has a valency of 1 (requires a sleeper).
  • A ditransitive verb like give has a valency of 3 (requires a giver, a gift, and a recipient).
  • The predicate's selectional preferences constrain the ontological category of each argument (e.g., the subject of sleep must be an animate entity).
02

Core Arguments vs. Adjuncts

A fundamental distinction exists between core arguments and peripheral adjuncts. Core arguments are essential participants required by the predicate's meaning to form a grammatically complete clause; adjuncts are optional modifiers adding circumstantial detail.

  • In Mary put the book on the shelf, the locative on the shelf is a core argument of put.
  • In Mary read the book on the shelf, the locative is an adjunct modifying the event.
  • Diagnostic tests like obligatoriness and entailment distinguish the two categories.
03

Thematic Role Assignment

Each argument is assigned a thematic role (theta-role) that captures its semantic relationship to the predicate. These generalized categories abstract over specific verbs to enable reasoning.

  • Agent: The volitional initiator of an action (John broke the window).
  • Patient: The entity undergoing a change of state (John broke the window).
  • Theme: The entity that is moved, perceived, or located (John gave the book to Mary).
  • Instrument: The means by which an action is performed (John broke the window with a hammer).
  • Goal/Recipient: The endpoint or receiver (John gave the book to Mary).
04

Syntactic Realization & Diathesis Alternations

The mapping between thematic roles and grammatical functions (subject, object) is not one-to-one. Diathesis alternations—such as the active/passive voice—rearrange this mapping while preserving core semantics.

  • Active: [Agent The chef] cooked [Patient the meal].
  • Passive: [Patient The meal] was cooked by [Agent the chef].
  • The linking theory component of a grammar specifies the rules governing which thematic role surfaces in which syntactic position for a given predicate.
05

Null Arguments & Implicit Participants

Arguments can be syntactically unrealized yet semantically present. Null arguments include pro-dropped subjects in languages like Spanish, and implicit objects in English.

  • In John ate, the patient argument (the food) is implicit but semantically understood.
  • Indefinite null complements (John ate something) differ from definite null complements (John noticed, where the stimulus is recoverable from context).
  • Proper interpretation of null arguments is critical for accurate information extraction and question answering.
06

Representation in PropBank & FrameNet

Two major resources formalize predicate-argument structures for computational use. PropBank annotates verb-specific numbered roles (Arg0, Arg1) mapped to prototypical thematic roles, while FrameNet defines semantic frames with frame-specific named elements.

  • PropBank: expect has Arg0 (expecter), Arg1 (thing expected).
  • FrameNet: The Commerce_buy frame evokes Buyer, Seller, Goods, and Money.
  • These annotations provide the supervised training data for modern neural Semantic Role Labeling systems.
PREDICATE-ARGUMENT STRUCTURE

Frequently Asked Questions

Clear, concise answers to the most common questions about the linguistic framework that maps sentences into their core semantic components—predicates and their arguments.

Predicate-argument structure is the linguistic framework that represents a sentence as a predicate (typically a verb) and its associated arguments—the noun phrases that define who did what to whom, when, where, and how. The predicate functions as the relational core, while arguments fill specific semantic roles such as Agent (the doer), Patient (the entity acted upon), and Instrument (the tool used). For example, in "Mary cut the cake with a knife," the predicate is "cut," with arguments Mary (Agent), the cake (Patient), and a knife (Instrument). This structure abstracts away from surface syntax to capture the underlying meaning, enabling machines to answer questions like "Who cut what?" regardless of whether the sentence is active or passive. Modern NLP systems derive these structures automatically using Semantic Role Labeling (SRL) models trained on annotated corpora like PropBank and FrameNet.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.