Who is most closely associated with the approach?

Roger Schank; an early parser paper was coauthored with Lawrence Tesler.

Does dependency here mean a software package dependency?

No. It refers to conceptual relationships in a meaning representation.

Can two paraphrases receive the same representation?

That is a central aim, when they express the same relevant meaning. Deciding that equivalence correctly is itself a substantive task.

AI Glossary / Natural language and speech / Advanced

Conceptual dependency

Definition

Conceptual dependency is a historical approach to representing the meaning of language through networks of concepts, primitive acts and their participants. Associated with Roger Schank, it aims to represent underlying events independently of the particular wording used to describe them.

What it means

Two sentences can differ grammatically while describing the same event. “Lena gave Arun a key” and “Arun received a key from Lena” emphasize different participants, yet can refer to one transfer. Conceptual dependency seeks a common conceptual representation beneath such variations.

Its dependencies concern meaning rather than simply grammatical links between words. It is therefore different from syntactic dependency parsing, which represents relations such as subject and object in a sentence. A conceptual representation can also make event structure and selected inferences explicit.

The language-independent representation is an aim of the theory, not a demonstrated guarantee for all languages and meanings. Its historical proposals about human memory should likewise be understood as theoretical claims rather than settled descriptions of how every person stores language.

How it works

An analysis identifies the relevant word senses and decomposes their content into conceptual structures. Participants fill roles such as actor, object and recipient. Relationships connect actions, states, time and other information, depending on the version of the framework.

Schank's primitive-act account uses ATRANS for changes in abstract relations such as possession or ownership, and PTRANS for physical changes of location. These distinguish moving a thing from changing who possesses it. More complex meanings can involve combinations of acts and relations rather than one primitive per word.

Mapping text into such a structure requires interpretation. The same verb can express different senses, and the input may omit details. An implementation must decide which inferences are licensed and which remain uncertain. A tidy representation can still be wrong if it fills a missing role with an unsupported guess or collapses two distinct events.

Three practical examples

A historical language-understanding system maps paraphrases to a shared event representation so they can support similar questions.
A teaching exercise distinguishes a change of ownership from moving an object between locations.
A modern knowledge-representation team studies the framework's treatment of roles and inference while choosing its own explicitly documented event schema.

Think of it this way

Think of converting different travel descriptions into a common itinerary with travelers, origins and destinations. The wording can vary while the represented journey stays the same.

The analogy illustrates normalization of selected meaning. An itinerary omits tone, perspective and many contextual details, just as any constrained conceptual representation may omit aspects that matter for another task.

Worked example

Separating ownership from location

In an invented scenario, Lena gives a spare key to Arun while both stand at the reception desk. A simplified conceptual sketch records an ATRANS of possession from Lena to Arun, with the key as the object. If she physically hands it over, a physical transfer can also be represented.

Now consider “Lena moved Arun's key from the desk to a drawer.” This describes a physical change of location, but it does not by itself say that ownership changed. Adding an ownership transfer would invent information.

Finally, “Arun received the key from Lena” could describe the same first event if the context establishes the same key, participants and time. It should not automatically merge with every other transfer involving those names.

These are explanatory sketches, not full canonical CD diagrams. Their purpose is to show why event identity and the kind of transfer must be modeled separately from surface wording.

Illustrative case study

A fictional archive project extracts equipment handovers from historical office notes. Its researchers study conceptual dependency to understand how a small collection of event types can connect differently worded descriptions.

They build a limited experimental representation for movement and possession changes, then test pairs of paraphrases and near-paraphrases. Reviewers inspect cases where a note says equipment was moved but does not say who owns it. Missing facts remain unspecified rather than being completed from a typical story.

The project treats the representation as a research design with a defined scope. It does not describe its new schema as a faithful implementation of every historical CD rule, or claim that it achieves general language understanding. The lesson is to make conceptual assumptions inspectable.

Common misconceptions and mistakes

Conceptual dependency is the same as a syntactic dependency tree. The two model different relations.
Every verb corresponds to one fixed primitive. Word senses and combinations of acts matter.
A language-independent target guarantees perfect translation. Mapping and coverage remain difficult.
A plausible inferred event is an observed fact. Inferences need an explicit evidential status.
The framework is a neural embedding method. Its classic representations are symbolic and structured.

Business use

The main business value for many readers is conceptual: distinguish how something is worded from the event being described. This can inform event extraction, duplicate detection and consistent record design, especially when ownership, movement and communication must not be confused.

For an actual implementation, specify the supported events, role definitions and inference rules. Preserve links to source text and evaluate near-miss cases that share vocabulary but differ in meaning. Do not assume that adopting a historical primitive vocabulary automatically makes a system complete, accurate or interoperable with other representations.

Check your understanding

Why should moving a borrowed projector into another room not automatically create an ownership-transfer event?

Answer: The statement establishes a location change. It does not establish a change in possession rights or ownership, so a separate abstract-transfer claim would need additional evidence.

Related terms

Semantic role
Semantic network
Event extraction
Semantic representation
primitive act
paraphrase
syntactic dependency

Learn next

Read the original primitive-act paper and compare its conceptual roles with syntactic roles. Create pairs of sentences that describe the same event and pairs that differ only in ownership or location, then identify what a representation must preserve.

Sources

Schank and Tesler, A Conceptual Dependency Parser for Natural Language: https://aclanthology.org/C69-0201.pdf. Early framework and conceptual-network account.

Roger Schank, The Primitive ACTs of Conceptual Dependency: https://aclanthology.org/T75-2008.pdf. Primitive acts and canonical-representation goals. The key-transfer sketches and archive case are original illustrations, not reproduced formal CD diagrams.

Previous: Conceptual data model Next: Conceptual graph