Nanos gigantium humeris insidentes!
You are currently browsing the Books category


  • July 29, 2010 2:40 pm


  • 1996-Building Probabilistic Models for natural Language
  • 2009-Statistical Language Models for Information Retrieval

Machine Learning(Book)

  • Pattern Recognition and Machine Learning
  • Semi-Supervised Learning

Relation Extraction(paper)

  • Extracting relations from Text: From Word Sequences to dependency Paths
  • Relation Extraction from wikipedia using subtree mining
  • Exploiting Syntactic and Semantic Information for Relation Extraction from Wikipedia
  • Semi-supervised Semantic Role Labeling Using the Latent Words Language Model
  • Kernel Methods for Relation Extraction

Prolog for natural language processing

  • July 11, 2010 2:29 pm

Chapter 3: The Linguistic Background

3.6 Functional Grammar and unification

This section is a good introduction to the functional structures research, and the last paragraph serves as a brief survey on this field.


  • July 9, 2010 10:25 am

Probabilistic linguistics / edited by Rens Bod, Jennifer Hay, and Stefanie Jannedy   P128.P73 P76 2003
Principles of knowledge representation / edited by Gerhard Brewka BD161 .B74 1996
Probabilistic Foundations of Reasoning with Conditionals / Judea Pearl, Moises Goldszmidt
Introduction to information retrieval / Christopher D. Manning, Prabhakar Raghavan, Hinrich Schütze QA76.9.T48 M26 2008  2008

Statistical language learning / Eugene Charniak   P98.5.S83 C47 1993

Statistical language models for information retrieval [electronic resource] / ChengXiang Zhai c2008 online

Reading Plan

  • July 6, 2010 5:46 pm


  • An introduction to English sentence structure / Andrew Radford
    • As a reference
  • English syntactic structures : functions and categories in sentence analysis / Flor Aarts and Jan Aar
    • Part II: Structure

Verb Syntax:

  • An empirical grammar of the English verb system / Dieter Mindt
    • The verb system
  • The grammar of English predicate complement constructions [by] Peter S. Rosenbaum       PE1380 .R6 c.2
    • The complement for predicate

Logic & Predicate

  • The semantic foundations of logic / Richard L. Epstein V.1       BC71 .E57 1994
  • Subject and predicate in logic and grammar / P.F. Strawson       B1667.S383 S83 2004

Book Record

  • July 6, 2010 1:52 pm


  • Natural language understanding
    • James Allen  QA76.7 .A44 1995
    • status: reading
    • comment: good for NLP
  • Principles of semantic networks : explorations in the representation of knowledge
    • contributors, Ale Q335 .P74 1991x
    • status: reading, almost finish
    • comment: Good start for knowledge representation
  • Graph-based knowledge representation [electronic resource] : computational foundations of conceptual
    • online c2009
    • status:reading
    • comment: Detail for Semantic Networks

  • Semantic networks in artificial intelligence
    • status:
    • edited by Fritz Lehmann ; general editor Ervin Y. Rodi Q335 .S423 1992x
  • Conceptual graphs for knowledge representation : First International Conference on Conceptual Struct
    • Q387 .I58 1993x
  • Knowledge representation : an AI perspective
    • Han Reichgelt  Q387 .R45 1991

English Language:


  • An introduction to English sentence structure / Andrew Radford
  • English syntactic structures : functions and categories in sentence analysis / Flor Aarts and Jan Aar **********

Verb Syntax:

  • An empirical grammar of the English verb system / Dieter Mindt
  • The grammar of English predicate complement constructions [by] Peter S. Rosenbaum       PE1380 .R6 c.2

Logic & Predicate

  • The semantic foundations of logic / Richard L. Epstein V.1       BC71 .E57 1994
  • Subject and predicate in logic and grammar / P.F. Strawson       B1667.S383 S83 2004

Common Logic Controlled English

  • July 4, 2010 1:51 am

Common Logic Controlled English

by sowa


  • July 2, 2010 11:15 pm

哦,那个书你没必要看的,上网搜一个图就可以了,输入rhichards rhetorical triangle就可以了,要书的话书名叫mencius on mind, I.A.Richards 写的
the meaning of meaning,那本里有一模一样的图


  • July 1, 2010 5:38 pm
  • Chapter 1 introduction
  • Chap. 2 basic conceptual graphs
  • Chap. 3 simple conceptual graphs
  • Chap. 4 set and first order logic semantics are the core chapters and should be considered as the basis.
  • For programming and algorithmic purposes the following materials can be added
  • to the base:
  • Chap. 6 basic algorithms for BG homomorphism
  • Chap. 5 relationships with constraint programming techniques
  • Chap. 7 techniques for trees and other tractable cases
  • Chap. 8 algorithms for other specialization/generalization operations, especially maximal joins
  • Chap. 10 rule processing), and Chap. 12 (algorithms for processing BGs with atomic negation
  • For modeling purposes the following parts can be added to the base:
  • Chap. 9 nested conceptual graphs ▲
  • Chap. 10 definition and use of rules ▲
  • Chap. 11 definition and use of constraints and their combination with rules▲
  • Chap. 12 definition and use of atomic negation
  • Chap. 13 semantic annotations

For more theory-oriented readers, expressivity, decidability and complexity results as well as stating equivalence with problems of other domains, are presented throughout the book, except in the last chapter.

This book is in electronic version, thus the note will be probably taken in the pdf file.

Chapter 13 The Evolving Technology of Classification-based Knowledge Representation Systems

  • July 1, 2010 12:33 pm

13.1 Introduction

The field of knowledge represnetation has not yet achieved the development of a general-purpose knowledge representation system(KRS)

This chapter outlines major principles that have been adopted by these systems, describes key features of several individual systems, and tries to identify unifying architectural themes among the various implementations that could evolve into a general purpose knowledge representation technology.

KL-ONE and its descendants share a commitment to the following architectureal features:

  • They are logic-based( i.e., they appeal to first-order logic for their semantics)
  • They draw a distinction between terminological and assertional knowledge, and each system implements its own specialized term-forming language
  • They include a classifier that organizes terms(concepts) into a taxonomy, based on subsumption relationships between terms

13.3 Classsification-based Technology

A classification-based KRS includes two languages:

  • a terminological language is designed to facilitate the construction of expressions that describe classes of individuals. — “a brown, three-legged dog”
  • an assertion language is to state constraints or facts that apply to a particular domain or world

assign B3LD to the concept representing “a brown, three-legged dog.” The assertion B3LD(Rover) states that Rover is a dog, is a brown thing, and has exactly three legs.

13.4 A KL-ONE family history

Important classifier technology issues

  • terminological languages
  • term classifiers
  • assertion languages
  • rule/constraint languages
  • truth maintenance
  • system performance

The defining characteristics of a TC system are

  1. its term-definition language
  2. its use of a concept classifier
  3. a database-like (object-centered) assertion language
  4. a forward-chaining recognizer
  5. a rule language
  6. automatic detection of inconsistent knowledge
  7. retraction of facts
  8. default reasoning

There is a variety of evidence suggesting that a KRS with an expressive terminological language(and incomplete classifier) is more useful tool than a KRS with a small terminological language( and a complete classifier).

Logical Dimensions of Some Graph Formalisms

  • July 1, 2010 9:44 am

Sowa’s CSs have already been applied in NLU, HGs have such a potential, and the three-level semantics is currently being implemented as a part of an NLU system.

From a logical point of view, knowledge represnetation systems are usually disscussed along the following dimensions:

  • Syntactic: What are the classes of  well-formed expressions in these formalisms? Is one class bigger than the other?
  • Semantic: What are their models? How can they be constructed? In other words, what are the entities the formalisms deal with, and what kind of relations can be expected to hold among them?
  • Proof theoretic: What kind of inferences do these formalisms support? What kind of syntactic structures can be built by applying the rules and can any graph be derived?
  • Decidability and computational complexity: How difficult is it to formally manipulate these objects? How difficult is it to prove something about them?
  • Whether these formmalisms use graphs in semantically, or just psychologically, significant ways?

12.2 Conceptual Graphs

The presence of graph-theoretic concepts in  the logic of conceptual graphs can be summarized as follows:

  • CGs represnet formulas/ propositions. A collection of CGs represnets a theory
  • since a model can be viewed as a collection of grounded formulas, one can represent a model as a collection of CGs, i.e., a theory
  • The inference rules can be rxpressed in graph-theoretic terms, like projection, restriction, and matching
  • But, since the theory of FOCGs corresponds to first-order logic, its semantics is standard and graphs can be eliminated without any loss in the expressive power.
  • Because of the above correspondence and the undecidability of satisfaction, there is no algorithm for deriving a model for a collection of GCs.
  • Models of a collection of CGs can be infinite
  • Graph properties of background knowledge, like proximity or connectedness, play almost no semantic role

Notice the conflict between the need for expressive power in order to handle natural language expressions and the difficulty in inferencing and constructing a model. Sowa’s work addresses only the first problem.

The trade-offs between the expressive poer and computability are discussed by Levesque and Branchmann, and their insights apply to graph-based formalisms,  too.


12.4 Graph Guide Inference in The Three-level Semantics

main ideas:

  1. Reasoning takes place in a three-level structure consisting of a referential level R(a partial order of theories representing background knowledge), an object level (describing current situation), and a metalevel (constraining types of models)
  2. The referential level is a collection of graphs whose nodes correspond to theories describing background knowledge, and whose links represent the partial orderings on these theories.
  3. An object-level theory T is augmented by theories from the referential level by extending paths from subformulas of T to corresponding theories of R. The resulting new, consistent, theories PT(T) can be then further extended to PT(PT(T)), and so on.
  4. Models of such extensions are then built, subject to metalevel constraints.

12.5 A Graph-based representation formalism?

  • First-order logic, especially in the form of conceptual graphs, makes certain types of knowledge readable. Until we are able to communicate with computers using natural language, first- order logic is perhaps the easiest formal language to carry conversations with them.
  • We have noticed some gaps in the theory of conceptual graphs:
  1. How to handle extensions of FOCGs?
    1. extracting relevant chunks from the knowledge soup to sonstruct a theory which makes deduction possible
  2. Lack of graph-based model theory.
    1. The problem is how to use conceptual graphs to respresent models of theory given as a set of CGs. Conceptual graphs would now have to be restricted in order to represent only grounded relations, functions, and terms.

Imagine a system that uses higraphs to represnet the structure of terms, and conceptual graphs for propositions ; in ference could be guided by topological properties of graphs representing knowledge, as it is in inheritance networkd with exceptions and the three-level semantics; models would be graphs, too.

    Page 1 of 212