[ CogSci Summaries home | UP | email ]
http://www.jimdavies.org/summaries/

Roger C. Schank (1972). Conceptual Dependency: {A} Theory of Natural Language Understanding, Cognitive Psychology, (3)4, 532-631

@Article{schank72,
  author =       "Roger C. Schank",
  key =          "Schank",
  year =         "1972",
  title =        "Conceptual Dependency: {A} Theory of Natural Language
                 Understanding",
  journal =      "Cognitive Psychology",
  pages =        "pages 532--631",
  volume =       "3",
  number =       "4",
}

Author of the summary: David Furcy, 1999, dfurcy@cc.gatech.edu

Cite this paper for:

SYSTEMS: one implementation mentioned but not named
The basis of natural language is conceptual.
The conceptual level is interlingual while the sentential level is language-specific.
Natural language understanding is largely predictive (i.e. based on expectations), thanks to the syntax and semantics of the conceptual level.
For NLU, syntax is only a pointer to the conceptual meaning and it is used at the same time as (interleaved with) conceptual processing.
Language cannot be studied without taking memory into consideration.

Summary

The main claim of this paper is that the "basis of natural language is conceptual." This conceptual base is interlingual, that is made up of concepts and relationships between concepts that are shared across languages. Understanding language means getting at that conceptual base. This is to be contrasted with the goal of most linguists which is to study whether specific sentences can be generated using formal methods essentially based on syntax. In Schank's view, "syntax ... is a pointer to semantic information rather than ... a first step to semantic analysis".

The Conceptual Dependency (CD) theory is proposed as a way of representing the information at the conceptual level. The main categories of concepts are PP's (i.e., picture producers, in other words, concrete nouns) and actions. Relations between concepts are dependencies. The main conceptualization of a clause is a two-way dependency between a PP (the actor) and an action. It is important to note that actions are broken down into primitive ACT's and do not correspond to verbs (the latter can be represented by several ACT's).

The syntax of the conceptual level is described by a set of rules which specify which type of concepts can depend on which other type, as well as the different kinds of dependency relationships between concepts. The semantics of the conceptual level determines which specific concepts can depend on other concepts based on the particular meaning of these concepts. There exists a dictionary of ACT's which specifies for each verb, its different meanings, and for each of the latter, its conceptual structure. In particular, each verb meaning requires a set of "conceptual cases" (among OBJECTIVE, RECIPIENT, DIRECTIVE, INSTRUMENTAL) and it is these rules that allow conceptual analysis to be largely guided by expectations about what should come next. In other words, the necessary cases constitute expectations to be confirmed. They can also be used to make inferences about details or about facts that are not mentioned because they are assumed to be known by the communicating parties.

At a higher level, conceptual relations indicate dependencies between conceptualizations, annotated with conceptual tenses (such as past, future, conditional, timeless). Note that these causal relationships will become central when Schank later represents higher level knowledge structures. For example, scripts are nothing but stereotyped causal chains. Other types of conceptual relations are the time and location of a conceptualization.

Towards the end of the paper, Schank emphasizes the difference between the meaning of a sentence (that is the conceptual content that CD is claimed to understand) and the meaning of the speaker (that is the intention of that content). This latter level of comprehension (interpretation or pragmatics) requires to consider the belief system of the understander, his/her world knowledge etc. In other words, NLU cannot be done without considering memory processes and content.

Summary author's notes:

In this summary, I have tried not to mention the details of CD because these have changed quite a bit over the years. For example, in this paper, Schank talks about a few classes of ACT's (like PACT, EACT, TACT, etc.). Later on, he will come up with eleven primitive ACT's (such as ATRANS, PTRANS, MBUILD etc.). But the motivation and the claims remain the same.

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:

JimDavies ( jim@jimdavies.org )

Last modified: Fri Oct 8 11:39:48 EDT 1999