Does the theory replace every dictionary sense with one entry?

No. It addresses productive relationships among interpretations; it does not imply that every meaning distinction disappears.

What does telic mean in qualia structure?

It concerns purpose or characteristic function. This usage should be distinguished from the related but separate discussion of telicity in event semantics.

Can a word have more than one contextually available reading?

Yes. A structured lexical analysis can permit alternatives, with additional context needed to prefer one.

AI Glossary / Natural language and speech / Intermediate

Generative lexicon

Definition

The generative lexicon is a theory of lexical meaning developed by James Pustejovsky. It represents words with structured semantic information and uses composition mechanisms to explain how they take on related interpretations in context. Generative here concerns the formation of interpretations, not the modern product category called generative AI.

What it means

A conventional dictionary can list several senses of a word. Generative lexicon theory asks whether some recurring differences can be explained through a shared structured representation instead of treating every contextual use as an unrelated entry.
For example, an object can be discussed in terms of its physical properties, what it is used for, or how it was created. Different surrounding expressions can draw on different aspects.
This is a linguistic framework, not a claim that dictionaries are unnecessary or that all ambiguity has one explanation. Its proposals concern systematic relationships among meanings and how those relationships participate in interpreting larger expressions.

How it works

The framework distinguishes several kinds of lexical information, including argument structure, event structure, qualia structure, and inheritance relations. Together they describe participants, event types, relevant properties, and links to other concepts.
Qualia structure organizes aspects such as what something consists of, what kind of thing it is, its purpose, and its origin. The conventional role names are constitutive, formal, telic, and agentive.
Composition combines this information with the requirements of surrounding words. Type coercion is one mechanism: an expression can be interpreted through an associated semantic type when its ordinary type does not directly satisfy the selecting expression.
An analysis can license more than one interpretation. Deciding which one a speaker intends can still require discourse and background knowledge beyond the lexical representation.

Three practical examples

A linguistics student analyzes why “start the report” can concern writing or reading, depending on the situation.
A semantic-resource designer records a tool's function separately from its material and manufacture, keeping different relations explicit.
An interpretation experiment compares an object's physical description with descriptions of the activity normally associated with it.

Think of it this way

Think of a word entry as a structured reference card with several connected tabs rather than a single flat label. The surrounding phrase can make one tab especially relevant. The analogy is limited because the theory includes formal composition constraints; interpretation is not simply a free choice among any facts someone happens to associate with an object.

Worked example

Consider the constructed sentence “Nadia started the report.” For a teaching analysis, suppose report is represented as an informational artifact associated with activities of writing and reading.
Step one: recognize that “started” invites an event interpretation. A report considered merely as an object is not itself an event.
Step two: inspect relevant relations associated with the noun. Writing concerns how the report is produced; reading concerns one ordinary use of it.
Step three: construct candidate interpretations: Nadia started writing the report, or Nadia started reading it. Both connect the object to an activity rather than asserting that the word report literally means an action in every context.
Step four: use context. “She had just opened a blank document” supports the writing interpretation, while “She reached page three before the meeting” supports reading.
This is an accessible illustration of the theory's motivation, not a complete formal derivation or a rule that restricts all possible contexts to two readings. It also shows why a lexical mechanism can provide candidates without independently settling speaker intention.

Illustrative case study

In a fictional project, Cedar Semantics designs a small resource for interpreting workplace requests. The team initially stores only broad object categories, which leaves phrases involving starting and finishing underspecified.
Researchers add separate relations for creation and ordinary use, then write test contexts that favor different interpretations. Reviewers retain ambiguous cases rather than forcing a single label when the surrounding conversation supplies too little evidence.
The imagined project demonstrates careful representation and evaluation. It does not claim that generative lexicon theory alone delivers a production assistant or that adding four fields resolves every expression.

Common misconceptions and mistakes

Generative lexicon theory is not a language model that generates text token by token.
The term qualia here names semantic explanatory roles; it should not be confused with claims about an AI system's conscious experience.
The agentive role concerns origin or creation and is not simply another name for the Agent thematic role in a sentence.
An object's purpose is not the only possible way someone can use it. Context can support unusual activities, and ordinary-use information should not be treated as an absolute behavioral law.
Systematic polysemy is not identical to every case of homonymy or accidental similarity between word forms.

Business use

For business language projects, the framework can help teams discuss what semantic information a domain resource should preserve. Distinguishing a document's content, production process, and use can make requirements more precise than a single broad category label.
Any implementation still needs explicit scope, annotation decisions, and testing on real requests. Treat the theory as a source of representational ideas whose usefulness must be evaluated, rather than a promise that a particular knowledge structure will automatically outperform other approaches.

Check your understanding

A system derives both reading and writing interpretations for “start the report.” Has it necessarily failed because it produced two candidates?

Answer: No. Both may be licensed before context is considered. The error would be treating one as certain without sufficient evidence, or failing to use context that clearly favors one.

Related terms

Word sense disambiguation
lexical semantics
polysemy
type coercion
qualia structure
compositionality
event structure
argument structure
semantic type

Learn next

distinguish polysemy from homonymy, then analyze several object-and-activity expressions with explicit contexts. Separate the interpretations a lexical account permits from the interpretation a particular conversation makes most plausible.

Activity expressions

Sources

Pustejovsky, “The Generative Lexicon,” original 1991 research paper: https://aclanthology.org/J91-4003.pdf. The paper introduces structured lexical representations and composition mechanisms. The report example and fictional project above are original teaching illustrations, not quoted experiments or a full implementation specification. Independent source review: Pustejovsky’s type-coercion definition, four qualia roles and novel/begin analysis (examples 38–46) were read.

Previous: Generative Language Interface Next: Generative model