Index of /~paritosh/research/Bote04/v1/carve/wfb-0.81

Name	Last modified	Size

Parent Directory		-
wfb-lifting.okbc	1998-02-19 19:19	38K
wfb.okbc	1998-02-19 19:17	262K
wfb-exceptions.kif	1998-02-19 19:17	326K
wfb-source.kif	1998-02-19 19:17	4.0M
wfb-knowledge.okbc	1998-02-19 19:17	692K
wfb-facts.okbc	1998-02-19 19:16	5.4M

World Fact Book Knowledge Base

The World Fact Book Knowledge Base (Release 0.81)

Adam Farquhar (axf@ksl.stanford.edu)
Richard Fikes (fikes@ksl.stanford.edu)
Gleb Frank (frank@ksl.stanford.edu)
Vilhelm Heiberg (heiberg@ksl.stanford.edu)
Sasha Zamler (zamler@ksl.stanford.edu)

Introduction

The World Fact Book project at the Stanford Knowledge Systems Laboratory is working to construct a substantial knowledge base covering basic geographic, economic, political, and demographic knowledge about the world's nations. Our goals in this project are to provide a useful knowledge resource, to explore ways for structuring large knowledge bases, and to provide a knowledge base large enough to stress existing knowledge representation systems.

The 0.8 Release (Valentine's Day release) is an early release, primarily for participants in DARPA's High Performance Knowledge Base (HPKB) program. We expect to issue several additional releases during March and April. These may include substantial modifications to the ontology as well as the base set of facts.

Point of Contact

Please contact Adam Farquhar or Gleb Frank. We welcome your suggestions and will be glad to help you make use of the KB.

Questions About the World Fact Book Knowledge Base

What is the WFB?

The CIA World Fact Book is a collection of facts of primarily geographical, economic and sociological nature about the countries and territories of the world. The Fact Book is distributed as a printed document, on CD-Rom, and via the web at http://www.odci.gov/cia/publications/nsolo/wfb-all.htm.

Although the web version is more up-to-date (it's the year 1996 version), we used the '95 CD-ROM version of the World Fact Book to construct the KB. The CD-ROM version is a structured SGML format, which makes parsing easier. The currency of information in the knowledge base wasn't our primary concern. We have been more interested in the task of representing the knowledge in the World Fact Book, rather than verifying the figures.

How was the knowledge extraction done?

The structure of this package pretty much reflects the process of knowledge extraction from the WFB. The original SGML file is structured as a set of records for each country; each record has a "field" string, an optional "subfield" string and a "data" string, e.g.

<field>Area</field>
<subfield>total area</subfield>
<data>337,030 sq km  </data>

The field and subfield strings are quite regular, and they correspond directly to the predicates in the parsed facts file. The data strings, in contrast, are highly irregular. This makes information extraction tricky and results in many special cases. Depending on the frequency of a case, it was either recognized and parsed in a special manner, or dumped into an "exceptions" file.

The facts file quite closely reflects the structure of the input file. The facts, with a few exceptions, are in the format

(<field> <country> [<subfield>] <data>...)

The <field>, <country and <subfield> are usually taken directly from the SGML structure. We did not apply sophisticated natural language techniques. When the structure of a data field was sufficiently regular, we tried to dissect the data strings into functional components, identifying elements of a list, value and measuring unit, etc.

How was the knowledge representation done?

We wanted to retain as much of original information from the World Fact Book as possible. Because of this, the representation was done "incrementally", i.e. the transformation from the source strings into the target frame-language structure was done in several steps, and the intermediate results were retained. The first step is the parsing and knowledge extraction. This step imposes little or no representational decisions on the structure of the facts file, which closely follows the predicate structure of the original file. The transformation into a high-level frame language was then done in a separate step. This allows easy conversion of the facts file into any representation language of appropriate level of expressivity. We used "lifting axioms", i.e. a set of rules, to map the predicates in the facts file to the ontology terms.

The ontology.

The aim for the World Fact Book ontology is to represent the facts in the WFB, allowing for easy browsing in an OKBC-based environment like Ontolingua, and providing machinery for reasoning in a first-order logic theorem prover like ATP. Some issues are covered more consistently and thoroughly than others. Topics connected with economy and population have received, overall, most coverage. This is in part due to the greater relevance of these topics for HPKB Challenge Problems, and in part because they contained more quantitative information and were easier to analyze.

The class hierarchy was linked to the HPKB upper level ontology, with the exception of the issues concerning measuring units and dimensions, which was linked to the Standard Units and Measures ontology in KSL.

What's in the package?

The package is currently distributed as a set of files specified in KIF and OKBC format. The files include:

Name:	wfb-facts.okbc
Description:	The file containing the basic facts of the WFB. The facts go in the order in which they were in the input file. To each fact a comment is appended indicating the corresponding line of the input file. The format of this and next file can be described as "OKBC-ized KIF", i.e. it's a bunch of facts in KIF format put inside an OKBC wrapper, which can be easily removed by deleting the OKBC header and footer, marked clearly in the file.
Format:	OKBC-ized KIF.
Example:	(age_structure afghanistan 0-14_years (* 42 %) );; line 588
Size:	5.2 MB, approx. 64000 facts.
Name:	wfb-knowledge.okbc
Description:	The file with class and name information on objects in the WFB. The sentences are in alphabetical order, duplicates are eliminated. For this reason, date information is also here; some dates tend to repeat dozens of times in the WFB (e.g., "1993"), and putting it into this file eliminates a large number of these redundant duplicates. It's also in "OKBC-ized KIF" format.
Format:	OKBC-ized KIF.
Example 1:	(commodity cereals );; line 82499
Example 2:	(pretty-name longs_wharf-jamaica "Longs Wharf");; line 56298
Example 3:	(date date_7_dec_1994 7 dec 1994 );; line 109653
Size:	700KB
Name:	wfb-source.kif
Description:	This file is intermediate between the real source file (in SGML format) and the parsed KIF files. Basically, it contains the source string for the parsing of arguments for each predicate in the facts file (with few exceptions for those cases when several predicates are produced from a single source line.)
Format:	KIF.
Example:	(source-text age_structure afghanistan 0-14_years "42% (female 4,342,218; male 4,507,141)");; line 588
Name:	wfb-exceptions.kif
Description:	Non-standard info from the WFB that didn't fit the common pattern syntactically. Each entry has a line number and a rough reason why the parser rejected it.
Format:	KIF.
Example:	(electricity \|Holy See (Vatican City)\| production "power supplied by Italy");; Line 49441:NothingToParse! "power supplied by Italy", by func.ParseNumber
Name:	wfb-onto.okbc
Description:	The World Fact Book ontology. This contains the definitions for the WFB classes, including taxonomies of some selected domains (like industrial sectors and natural resources), machinery to operate with some concepts frequently occurring in the WFB, like fractions, and links between the concepts in the WFB and the HPKB upper level ontology. The ontology Physical-Quantities has also been intensively used.
Format:	OKBC.
Name:	wfb-lifting.okbc
Description:	The KIF axioms that link the predicate in the facts file to a corresponding frame-language. Could be used to answer queries formulated in the terms of the WFB ontology or, if run in forward-chaining mode, to produce frame-language statements from the WFB facts. The WFB facts are assumed to be in LISP package WFB.
Format:	OKBC-ized KIF.
Example:	(=> (wfb::population ?georef ?number) (exists ?population (and (set-cardinality ?population ?number) (country-population ?population) (population ?georef ?population))))

History.

Feb. 14 1998. Version 0.8 ("Valentine's Day Release").

Feb. 19 1998. Version 0.81. Fixed some bugs; changed format of most files from KIF to OKBC-ized KIF.