On The System Of Concepts Relations And Outline Of The Natural Language System

Sho Yoshida

In this 1975 paper, Sho Yoshida of the Kyushu Institute of Technology proposed a pioneering framework for a Natural Language System (NLS) designed to simulate human-like thinking. Unlike contemporary systems focused on narrow syntactic parsing, Yoshida’s model centers on a "Desire System," suggesting that true intelligence arises from goal-driven behavior where language is a tool to satisfy internal needs.

Introduction: The Need for Intentionality in AI

During the mid-1970s, natural language processing (NLP) was largely restricted to database queries and structural analysis. Yoshida identified a critical missing component: the "will" to communicate. He argued that thinking and behavior are fundamentally motivated by the need to satisfy internal desires. Consequently, a robust NLS must integrate an internal drive with a comprehensive knowledge of the world to determine appropriate actions based on linguistic input.

Architecture of the Thinking System

Yoshida’s proposed "Thinking System" operates through a multi-stage process that maps external stimuli to internal goals.

Figure 1: Block diagram of the proposed Thinking System architecture, highlighting the central role of the Desire System and Processor.

The architecture consists of four primary components:

  1. Input Recognition: Sentences are decomposed into bunsetsu (Japanese grammatical units) to analyze dependencies.
  2. Normal Form Generation: Natural language is translated into a formalized predicate-logic structure, capturing "case relations" (e.g., agent, object, action).
  3. System of Concepts Relations: This serves as the core knowledge base, categorizing words into abstract concepts, concrete nouns, and verbs with defined semantic roles.
  4. The Desire System: Acting as the controller, this system monitors inputs for triggers related to internal needs (e.g., "appetite" or "evasion") and generates "what to do" responses.

Knowledge Representation and Semantic Inference

A key innovation in the paper is the "System of Concepts Relations," an early precursor to modern knowledge graphs. This system organizes information through various "view-points." For instance, a single object like a "cup" can be categorized by its material (china) or its purpose (drinking).

Figure 5: A partial classification of concrete objects and basic words, illustrating the hierarchical organization of knowledge.

By traversing these relational graphs, the system performs semantic inference. For instance, it can conclude that a "cup" is a "concrete object" that is "breakable" based on its hierarchical properties. If the system knows the action "dropping" leads to "breaking," it can predict the consequences of an action before it occurs. This causal chaining allows the system to evaluate the world dynamically.

Desire as the Engine of Cognition

The most striking aspect of Yoshida’s work is the assertion that "desire is the basic cause of thinking." The system does not merely react to text; it evaluates information based on its internal state. For example, the concept of "running" might be linked to a desire for "evasion" through a chain of reasoning: run -> move -> tired -> evasion. This allows the AI to generate behaviorally relevant internal states and responses.

Conclusion and Future Challenges

While Yoshida provided a comprehensive blueprint, he acknowledged significant hurdles. The logic for "view-point selection"—how the system chooses the relevant context for an object—remained an open question. Additionally, scaling the simple desire models to the complexity of human psychology presented a massive engineering challenge. Despite these limitations, the paper remains a foundational vision for goal-oriented, knowledge-rich AI.

Summary updated

BibTeX
@inproceedings{yoshida1975system,
  author = {Sho Yoshida},
  title = {{On The System Of Concepts Relations And Outline Of The Natural Language System}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1975},
}
Figure