Long before search bars and AI chatbots, librarians and computer scientists were already asking a difficult question: could a machine think like an expert reference librarian? The answer took shape in a family of early experimental tools called expert systems. These systems tried to capture the reasoning of a human specialist and apply it to the messy task of finding the right information. Among the most discussed examples in information science courses are COMIT, REFSEARCH, RESEDA, CANSEARCH, and PLEXUS. Each tackled a different problem, but together they explain how intelligent information retrieval first began to take form.

Table of Contents

What expert systems mean in information retrieval

An expert system is a type of artificial intelligence that stores the knowledge of a human specialist and uses it to solve problems. It usually has two main parts: a knowledge base, which holds facts and rules, and an inference engine, which applies those rules to reach conclusions. In information retrieval, the “expert” being copied is often a skilled reference librarian or search intermediary who knows how to turn a vague question into a precise search.

This matters because ordinary retrieval depends heavily on the user knowing exactly what to type. A patron searching a database may not know the correct subject headings, synonyms, or search commands. Expert systems were designed to close that gap by guiding the search, reformulating queries, and reasoning about what the user actually needs. They were largely experimental prototypes built in the 1970s and 1980s, but their ideas still shape modern intelligent search.

COMIT and the rise of pattern matching

COMIT is one of the earliest names linked to intelligent text processing. It was developed at the Massachusetts Institute of Technology (MIT) by Victor Yngve and his collaborators, first appearing in 1957. COMIT is recognised as the first true string-processing language, and it was built around a powerful idea: pattern matching. Instead of treating text as fixed words to be matched letter for letter, COMIT could search for structures and patterns inside the text.

This was a major step forward. Traditional retrieval relied on exact keyword matches, so a small difference in wording could cause a search to fail. Pattern matching allowed a system to compare a structured query against descriptive records and find results that fit the pattern, even when the exact words differed. Because of this, COMIT influenced later languages such as SNOBOL, which also treated patterns as a core feature.

Why pattern matching mattered for retrieval

In an information retrieval context, the value of COMIT lies in the technique rather than a single application. When a researcher wants details about a person, they may supply a structured query made of attributes like name, profession, and time period. A pattern-matching approach can compare that structured query against descriptive or biographical records and surface the closest matches. The pattern-matching paradigm that COMIT pioneered became a building block for later systems that needed to match queries against complex records rather than plain word lists.

REFSEARCH and support for reference work

While COMIT was about how machines handle text, REFSEARCH focused on a very practical library task: helping people find the right reference source. REFSEARCH is described as a system that recommends the appropriate sources a user should consult to answer a particular type of question. Rather than retrieving the final answer directly, it points the user toward the reference works most likely to contain it.

This makes REFSEARCH interesting for two audiences. For students, it works as a teaching tool that builds reference skills by showing which kind of source suits which kind of question. For practising reference librarians and information specialists, it acts as a computerised assistant that supports the everyday job of matching a query to the correct dictionary, directory, encyclopaedia, or database.

Reference work is one of the oldest skills in librarianship. A good reference librarian listens to a question, clarifies what is really being asked, and then guides the user to the best resource. REFSEARCH attempted to encode part of that judgement into software. For LIS students, it is a clear example of how expert systems were aimed not at replacing the librarian, but at supporting and teaching the reference process.

RESEDA and reasoning over biographical data

RESEDA shows how expert systems can move beyond simple matching into actual reasoning. It was a research project based at France’s Centre National de la Recherche Scientifique (CNRS), built to manage factual databases of biographical data. The information was described using a special knowledge representation language, sometimes called a “meta-language,” based on a case-grammar approach drawn from artificial intelligence.

What makes RESEDA notable is how it answered questions. When a user posed a formal question, the system first tried to find a direct match in its database. If no direct answer existed, RESEDA did not simply give up. Instead it applied inference procedures to reason its way toward an answer. One type, called a “transformation,” reshaped the query to find related data. Another, based on “hypotheses,” could even propose new causal links between events stored in the base, effectively creating information that was not explicitly recorded before.

Because its database held biographical and historical records, RESEDA is often presented as an early example of applying expert-system reasoning to a specialised humanities domain. It demonstrated that retrieval could involve genuine inference, not just lookup, which was a striking idea for its time.

CANSEARCH and expert searching in medicine

CANSEARCH is one of the best-documented expert systems in this group, and it targeted a high-stakes field: cancer therapy. It was developed by A. Steven Pollitt at Huddersfield Polytechnic in the UK, designed to let doctors directly search cancer-therapy literature stored in the MEDLINE database. The central goal was bold for the 1980s: allow end users to run good searches without being trained in the underlying retrieval system.

To achieve this, CANSEARCH used a principle called search space abstraction. Users made selections from a touch terminal using simple menus, and a rule-based program written in PROLOG handled the expert reasoning behind the scenes. A modified subset of the Medical Subject Headings (MeSH) was stored on a microcomputer so the system could translate a doctor’s menu choices into a proper search query. In effect, CANSEARCH packaged the skill of a trained MEDLARS search intermediary into software a clinician could use directly.

The project performed well enough to justify the approach, and it later influenced a whole line of interface research. Some features of CANSEARCH survived into later “view-based searching” tools, showing how an experimental prototype can shape long-term thinking in information science.

PLEXUS and other referral systems

No discussion of these systems is complete without PLEXUS, an expert system built as a referral tool for public libraries. Developed by Alina Vickery and colleagues at the University of London with British Library funding, PLEXUS aimed to perform the same kind of task as a human reference librarian, and to do it in a way that would look intelligent if a person did it.

The prototype was deliberately restricted to a single subject, gardening, because the topic had broad appeal among public library users. PLEXUS would obtain a description of the user’s problem, expand it by deriving extra concepts or asking follow-up questions, and then turn that into a search strategy applied to referral resources. If the first attempt failed, the system reformulated the strategy until it produced a useful result. This made PLEXUS a clear model of how knowledge bases, semantic categories, and inference rules could imitate reference judgement.

PLEXUS sits alongside other advisory tools such as POINTER and Online Reference Assistance, all of which tried to help users locate reference sources and factual data. Together with RESEDA and CANSEARCH, they form a generation of knowledge-based retrieval systems that proved expert-system techniques could be applied across many subject domains.

Why these systems still matter for LIS students

It is easy to dismiss these prototypes as outdated, since most were experimental and never reached wide commercial use. But their importance is conceptual. They established that information retrieval could involve reasoning, not just matching, and that the expertise of a librarian could, at least partly, be encoded into rules. Pattern matching, query reformulation, knowledge bases, and domain-specific search strategies all trace back to this period.

For students studying information processing and retrieval, these systems offer a clear timeline. COMIT introduced pattern matching; REFSEARCH and PLEXUS modelled reference and referral work; RESEDA showed inference over structured data; and CANSEARCH proved end users could search expertly with the right interface. The intelligent search tools and recommender systems used today are direct descendants of these early experiments, which is why they remain a standard part of the syllabus.

What do you think? If an expert system could capture the reasoning of a skilled reference librarian, which parts of that human judgement do you think are easiest to encode as rules, and which parts may always need a human touch?

How useful was this post?

Click on a star to rate it!

Average rating 0 / 5. Vote count: 0

No votes so far! Be the first to rate this post.

We are sorry that this post was not useful for you!

Let us improve this post!

Tell us how we can improve this post?

References
  1. https://www.emerald.com/insight/content/doi/10.1108/eb026798/full/html
  2. https://en.wikipedia.org/wiki/COMIT
  3. https://en.wikipedia.org/wiki/SNOBOL
  4. https://lucidea.com/blog/artificial-intelligence-technology-trends-that-arent-out-there-anymore/
  5. https://www.sciencedirect.com/science/article/abs/pii/S0020737384800073
  6. https://dl.acm.org/doi/abs/10.1145/511793.511821
  7. https://www.sciencedirect.com/science/article/abs/pii/0306457387900537
  8. https://www.oclc.org/research/events/dss/pollitt.html
  9. https://eric.ed.gov/?id=EJ351271
  10. https://www.sciencedirect.com/science/article/abs/pii/0306457387900525
  11. https://eric.ed.gov/?id=EJ421695

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

Information Processing & Retrieval

1 Intellectual Organisation of Information

  1. Intellectual Organisation of Information
  2. Meaning of Intellectual Organisation of Information
  3. Why IOI is Necessary?
  4. IOI in Indexing Systems
  5. IOI and Indexing Languages
  6. IOI in User Services
  7. IOI and Content Analysis
  8. Information Retrieval Systems – Changing Environment
  9. Future Trends

2 Indexing Languages–Part I – Concepts and Types, Subject Headings Lists and Thesauri

  1. Indexing and its Types
  2. Indexing Language
  3. Vocabulary Control
  4. Classification Schemes
  5. Subject Headings Lists
  6. Thesaurus
  7. Thesaurofacet
  8. Classaurus
  9. Sears List of Subject Headings
  10. Library of Congress List of Subject Headings

3 Indexing Languages–Part II- Classification Schemes

  1. Dewey Decimal Classification (DDC) Scheme
  2. Universal Decimal Classification (UDC) Scheme
  3. Library of Congress Classification (LCC) Scheme
  4. Colon Classification (CC) Scheme
  5. Bibliographic Classification (BC) Scheme
  6. Library Bibliographical Classification (BBK) Scheme
  7. Broad System of Ordering (BSO) Scheme
  8. Special Classification Systems

4 Indexing Systems and Techniques

  1. Indexing Principles and Process
  2. Pre-Coordinate Indexing Systems
  3. Post-Coordinate Indexing Systems
  4. Automatic Indexing
  5. Non-Conventional Indexing: Citation Indexing
  6. Web Indexing

5 Evaluation of Indexing Systems

  1. Purpose of Evaluation
  2. Levels of Evaluation
  3. Evaluation Criteria
  4. Recall and Precision
  5. Other Performance Measures
  6. Relevance
  7. Evaluation Methodology
  8. Evaluation Experiments

6 Principles and Evolution of Bibliographic Description

  1. Bibliographic Description: An Overview
  2. Scope and Objectives of Bibliographic Description
  3. Evolution of Bibliographic Description
  4. Ranganathan’s Principles
  5. ISBDs
  6. Bibliographic Formats
  7. Electronic Resource Description
  8. Models of Bibliographic Description
  9. Bibliographic Description: Entities, Attributes and Relationships

7 Rules for Bibliographic Description

  1. Bibliographic Description: Its Origin
  2. Development of Anglo-American Code
  3. The International Standard Bibliographic Description (ISBD)
  4. Impact of ISBD on Catalogue Codes
  5. Bibliographic Description for Non-Print Materials
  6. Guidelines for Bibliographic Description of Electronic Resources
  7. Guidelines for Bibliographic Description of Internet Resources
  8. Rules for Description of Electronic Resources in AACR2 Revision 2002

8 Standards for Bibliographic Record Format

  1. International Standard Bibliographic Description (ISBD)
  2. MARC Format
  3. UNIMARC
  4. Common Communication Format (CCF)
  5. Indian Standard

9 Metadata- MARC21-856 Field, Dublin Core, TEI

  1. MARC21 – 856 Field
  2. Dublin Core Metadata Initiative (DCMI)
  3. Text Encoding Initiative (TEI)
  4. Procedure of Electronic Resource Description

10 Norms and Guidelines for Content Development

  1. Introduction
  2. Needs and Guidelines
  3. Standards Related to Electronic Content
  4. W3C Recommendations
  5. Electronic Text Encoding and Interchange
  6. Dynamic Content

11 Introduction to HTML and XML

  1. World Wide Web and Markup Languages
  2. Standard Generalized Markup Language (SGML)
  3. HyperText Markup Language (HTML)
  4. Basic HTML Tags
  5. Linking
  6. URLs
  7. HTML and the Browser
  8. eXtensible Markup Language (XML)
  9. XML Syntax and Semantic Tags
  10. Document Type Definition (DTD)
  11. Implications of XML in Library and Information Activities

12 Web-based Content Development

  1. What can be done with World Wide Web?
  2. Hypertext, Hyperlink, and Hypermedia
  3. Hypertext Markup Language (HTML)
  4. Introduction to Dynamic HTML
  5. Web Interface to Database Linking
  6. Introduction to XML
  7. XML Document Design
  8. Multimedia Web Resources
  9. Web Servers
  10. Website Hosting
  11. Tools for Web Page Designing

13 Multilingual Content Development (Using Unicode)

  1. Character Representation in Computer
  2. American Standard Code for Information Interchange (ASCII)
  3. Indian Scenario and Indian Standard Code for Information Interchange (ISCII)
  4. UNICODE
  5. Web Content Development Through UNICODE
  6. Applications of UNICODE
  7. Applying UNICODE to the Libraries
  8. Problems Associated with UNICODE

14 ISAR Systems- Objectives, Types, Operations and Design

  1. Users and Their Information Needs
  2. Objectives of ISAR Systems
  3. Types of ISAR Systems
  4. Design of ISAR Systems
  5. Evaluation of ISAR Systems

15 Compatibility of ISAR Systems

  1. Need for Compatibility Among ISAR Systems
  2. Scope of Compatibility in ISAR Systems
  3. Areas of Compatibilities in ISAR Systems
  4. Principal Issues of Compatibility in ISAR Systems
  5. Compatibility of Online IR Systems
  6. Approaches Towards Compatibility in ISAR
  7. Quality Control and Compatibility

16 Intelligent Information Retrieval Systems

  1. Introduction
  2. Expert Systems
  3. Expert Systems for Information Processing and Retrieval
  4. Components of Expert Systems
  5. Knowledge Representation
  6. Knowledge Engineering
  7. Artificial Intelligence Based Decision Support Systems (DSS)
  8. Pattern Recognition

17 Information Retrieval Processes and Techniques

  1. Information Retrieval Systems
  2. Databases
  3. Information Retrieval Systems: Purpose, Components, and Functions
  4. Indexing and Information Representation
  5. Vocabulary Control
  6. Searching
  7. Information Seeking and User Interfaces
  8. Web Information Retrieval Systems
  9. Intelligent Information Retrieval

18 Information Retrieval Models and Their Applications

  1. Information Retrieval
  2. Information Retrieval Techniques
  3. Models Based on Input/Output
  4. Models Based on Theories and Tools

19 Search Strategies, Processes and Techinques

  1. Search File – An Essential Component
  2. Search Strategies and Pre-requisites
  3. Search Techniques
  4. The Information Search Process
  5. Online Searching
  6. How the Search Engines Work
  7. Common Search and Retrieval Features of Web Search Engines