Language

1
Language

• A system of symbols and rules used to communicate
  knowledge
• Between humans (e.g., Natural Language)
• Between machines (e.g., Binary Code)
• Between human and machine (e.g., Programming
  Language, GUI)
• etc.

2
Two Flavors of Knowledge

• Explicit
• Facts (Potentially Infinite)
• No computation required
• Communication can be prohibitively inefficient
• E.g., Table of multiplication facts rather than a
  procedure
• Implicit
• Finite Facts Computation Procedure
• Computation procedure is finite and explicit
• Communication can become very efficient
• Concise representations allows for communication
  (computations as needed) potentially infinite
  knowledge

3
Logic as Language

• Origins are Pre-Aristotle
• A language for the explicit communication of
  ones knowledge between humans
• Explicit Facts General purpose inference
  procedure
• Frege (1879) develops modern symbolic logic, the
  predicate calculus, complete with proof theory -
  a formal languagefor pure thought
• Exploited by mathematicians and philosophers
  demanding tools for rigorous expression and
  formal, unambiguous human communication
• Becomes the foundation for mathematical
  reasoning.
• Principal developmental constraint Human
  cognitive processes

4
The Role of Computers Communication

• Store and communicate knowledge between humans
• Use finitely expressed procedures to infer
  communicate a potentially infinite amount of
  otherwise implicit knowledge
• Like a Book?
• Can afford MORE EFFICIENT COMMUNICATION
• Like a Human?
• Can in many cases compute faster and more
  accurately.

5
The Role of Communication In Computing

• Step 1 Formulation - Human Focus
• Human communicates with Human
• Step 2 Program - Machine Focus
• Human conforms and communicates with the computer
  
• Step 3 Application - Human Focus
• Computer conforms and communicates with Human

6
Step 1 Formulation

• Human prepares to communicate with machine by
  formulating knowledge.
• Typically communicates with other humans for
  accuracy and validation.
• Logic is a classic and fundamental language.
• Identify Goal
• body of knowledge and
• form of human-machine communication for step 3
  (input/output)
• Identify base knowledge
• explicit facts required to derive body of
  knowledge
• Identify computation procedure
• Explicit procedure algorithm used to derive
  subset of knowledge representing output given
  input and base knowledge

7
Steps 2 and 3

• Step 2 Program
• Human conforms and communicates with machine
• Human reformulates in a language the computer can
  process
• Programming language becomes means of expression
  for human to communicate knowledge and
  computation procedure to machine.
• Step 3 Application
• Machine comforts to communicate with Human
• Human user queries/dialogs with computer
  providing input
• Computer executes program, derives, forms and
  communicates response to human

8
Programming Languages

• Origins -- 1940s
• Developed for Two Purposes
• For Controlling the operations of a computer AND
• For humans to express knowledge
• Conflicting Developmental Constraints
• The con Neumann architecture
• Human cognitive processes
• Knowledge formulation in logic is not sufficient
• Must program the machine
• The program becomes the persistent artifact of
  communication not the human formulation

9
Bridging the Gap
Objects and Attributes Automatic Inference
(Inheritance) Automatic Memory Management
Assembler
C
ALGOL Family
Functional Family
Logic Programming
01011 01010
Programming Languages Bridging the Gap
Logic
Designed for Formal Expression and Communication
of Human Knowledge
Designed to operate a machine designed to add
binary numbers
10
Logic as Program

• Program
• set of axioms
• declarative logic statement of knowledge
• Computation
• Constructive proof of goal statement from a
  program
• Not part of program -- an abstract machine --
  like the accumulator in a VN
• Answer natural logic consequent of the program
• Original Motivation
• Desire to automate the proof procedure of
  predicate calculus so that theorems could be
  automatically generated by the computer
• In general, however, this is not possible (proof
  by Godel)
• Positive Impact
• For a class of interesting problems a limited
  proof procedure can be useful
• A declarative logic-based language can afford
  precise logical syntax and semantics based on
  general-purpose and natural computation
  procedure
• A classic logic-based expression is transformed
  into effective computation

11
Logic Programming Historical Highlights

• 1965 Robinsons Unification Algorithm and
  Resolution Principle mechanizes proof procedure
• 1970s
• Kowaski formulates a procedural interpretation of
  Horn Clause Logic. Shows that an axiom of the
  form
• A if B1 and B2 and BN
• Can be interpreted and executed as a procedure of
  a recursive program based on resolution proof
  procedure with unification algorithm.
• Colmerauer develops Prolog based on Kowlaskis
  procedural interpretation
• vanEmden and Kowalski develop formal semantics
  for Logic Programming
• Warren dispels myths of impracticability with an
  efficient implementation based on easily
  programmed abstract machine (the WAM)
• 1980s -
• Advanced techniques execute more logical specs
  as efficient programs
• Constraint Logic Programming as a generalization
  of LP allows for a framework that maintains
  declarative logic syntax and semantics while
  solving classes of CSPs as efficiently as
  theoretically possible.

12
The Ideal and The Reality

• Ideal
• A problem decoratively stated in symbolic logic
  executes on a vonNeuman machine as efficient
  program
• Benefits
• Programming becomes classic logical specification
  and inherits the known benefits of formal logic
  as language (precise semantics and natural
  inference mechanism)
• Intent and correctness of programs are more
  easily accessible, understood and validated by
  humans (and by machines)
• Expression/Communication of domain/problem
  knowledge is not cluttered with knowledge about
  the machine or about the idiosyncrasies of
  managing machine resources.
• Reality
• Currently not ALL clear, elegant logic
  specifications in LP execute as efficient logic
  programs
• None the less LP allows for specialized
  algorithms to be expressed as declarative logic
  programs.
• Research in LP is focused on broaden the class of
  specs that do execute as efficient logic programs

13
LPs Role in this Course

• Course focus
• AI and Knowledge Representation and Automated
  Reasoning
• How computers can help HUMANS capture, compute
  and communicate knowledge.
• Logic Programming
• Has close ties to formal logic - the tried and
  true foundation for the formal representation and
  communication of human knowledge
• Provides a programming foundation for formally
  representing knowledge and inference
• Allows us, as software engineers, to write
  programs ignoring machine issues (e.g., memory
  structure and management) and to concentrate on
  knowledge representation, human reasoning and
  formal human communication.

14
Why Applications in Document Generation

• Encompass a broad range of AI topics and
  techniques
• Documents are themselves human communication
  artifacts
• They are produced by some process
• The represent and communicate knowledge through
  language
• The act of authoring documents can range
• From the very creative (fictional literature)
• To the very unimaginative and simplistic
  (filling in a form)
• And include things in between like finding,
  weaving together and refining previous written
  text to form a new document
• Exploring how we might model the human processes
  involved in document authoring requires ideas
  about how we might represent and reason about
  knowledge, language and the human creative
  processes.
• Great foundation for exploring the potential and
  the limits of the role of computers in human
  creativity and communication

15
Course Overview

• Logical Foundations
• Properties and Mechanics of Classic Symbolic
  Logic
• Logic Programming and PROLOG
• Practical tool kit for implementing
  knowledge-based systems
• Knowledge Representation
• Basic techniques for modeling common-sense
  domains
• Three Applications
• Brutus -- Creative Writing
• DocCon -- Business document authoring, knowledge
  management and reuse
• Test Generation

16
Brutus

• Background
• Academic research in automatic story generation
• Designed to explore computational models of human
  creative authorship
• Developed in Prolog (FLEX)
• Focus
• What kinds of knowledge is represented
• What kinds of processes are modeled
• What kinds of automatic reasoning techniques are
  used
• How is natural language used
• Students will have an opportunity to explore and
  extend the code

17
DocCon

• Background
• Commercial System developed at an RPI incubator
  company, Document Development Corporation, in
  collaboration with IBM Research
• Designed to assist in managing and reusing
  document knowledge to assist in authoring
  business documents
• Carefully engineered and developed a variety of
  languages (Java, C, Delphi)
• Focus
• Business Problem
• System architecture
• Document Knowledge Representation Scheme
• Interactive Document Configuration
• Students will have an opportunity to develop
  authoring solutions for a class of documents
  using DocCon

18
Transitive Closure Version 0

• -------------------------------------------------
  -------------
• Version 0
• - Direct Rep of logic
• - Does not terminate (Try - closure(1,3) for
  test0)
• - Inefficient
• -------------------------------------------------
  -------------
• closure0(X,Y) - base(X,Y). arcs in graph
• closure0(X,Z) - closure0(X,Y),
  closure0(Y,Z). transitive step

19
Transitive Closure Version 1

• -------------------------------------------------
  -------------
• Version 1
• - A little less Direct Rep of logic
• - Terminates for acyclic graphs
• - works on test0 but not on
  test1.
• - Still Inefficient
• -------------------------------------------------
  -------------
• closure1(X,Y) - base(X,Y).
• closure1(X,Z) - base(X,Y), closure1(Y,Z).

20
Transitive Closure Version 2

• -------------------------------------------------
  -------------
• Version 2
• - A little less direct rep of logic
• - Terminates for cyclic and acyclic graphs
• - skips visited nodes
• - works on test0 and test1
• - Inefficient - Traverses all paths when
  only need one.
• -------------------------------------------------
  -------------
• closure2(X,Y) -
• closure2a(X,Y,X).
• closure2a(X,Y,_) - base(X,Y).
• closure2a(X,Z,Avoiding) -
• base(X,Y),
• not(on(Y,Avoiding)),
• closure2a(Y,Z,Y Avoiding).

21
Transitive Closure Version 3

• -------------------------------------------------
  -------------
• Version 3
• - Warshalls algorithm for transitive closure
  in Prolog
• - Indirect Rep of logic
• - Terminates for cyclic and acyclic graphs
• - Efficient (Does not visit unnecessary
  paths)
• -------------------------------------------------
  -------------
• "Very often you will be tempted to think that
  it is time
• to abandon Prolog for C. Often you will do
  better to replace
• a specification (in Prolog) by a program (in
  Prolog)."
• Good algorithms are important no matter what
  programming Language.