ThoughtTreasure is a commonsense knowledge base and architecture for natural language processing. It contains both declarative and procedural knowledge.
Contents |
ThoughtTreasure's knowledge base consists of concepts, which are linked to one another by assertions. An assertion is represented in the form
@timestamp:timestamp|[concept ...]
Some examples of assertions in ThoughtTreasure are:
[isa soda drink] (A soda is a drink.) [part-of phone-ringer phone] (A phone ringer is part of a phone.) [green green-pea] (A green pea is green.) [diameter-of green-pea .25in] (The diameter of a green pea is .25 inches.) [duration attend-play NUMBER:second:10800] (The duration of a play is 10,800 seconds.) [product-of Intel-8080 Intel] (An Intel 8080 is a product of Intel.) @19770120:19810120|[President-of country-USA Jimmy-Carter] (Jimmy Carter was the President of the USA from January 20, 1977 to January 20, 1981.)
ThoughtTreasure contains a total of 27,000 concepts and 51,000 assertions. It has an upper ontology and several domain-specific lower ontologies such as for clothing, food, and music.
Each concept is associated with zero or more lexical entries (words and phrases). Two languages are supported: English and French. ThoughtTreasure has 35,000 English lexical entries and 21,000 French lexical entries. In addition to open-class lexical entries such as nouns, verbs, adjectives, and adverbs, ThoughtTreasure also contains closed-class lexical entries such as conjunctions, determiners, interjections, prepositions, and pronouns. It also contains a dictionary of names.
Zero or more features are attached to each lexical entry. There are 118 features [1]. Examples are ZEROART (zero article taker), SING (singular), FML (formal), CAN (Canadian), ENG (English), and N (noun). Argument structure is provided for verbs. For example, the argument structure for the concept walk-into is
*> S ---- (from IO[2]) into IO
ThoughtTreasure contains 93 scripts, or representations of typical activities.
ThoughtTreasure contains 29 grids, which represent the arrangement of objects in typical locations such as hotel rooms, kitchens, and theaters. Grids are connected together by wormholes.
ThoughtTreasure includes a planning agency for achieving goals in a simulated world and an understanding agency for understanding stories and asking and answering questions.
ThoughtTreasure contains the following procedures for natural language processing:
ThoughtTreasure contains the following procedures that deal with space:
It contains operations dealing with parts and wholes of objects, grids (distance, subspace), large space (planetary distance, polity containment), and nested space (room, floor, building, city, planet).
Other procedures in ThoughtTreasure include:
ThoughtTreasure can be used to add common sense to applications by using its knowledge base or by communicating with a ThoughtTreasure server.
The ThoughtTreasure knowledge base is available in three file formats:
Python code for accessing TTKB is available. TTKB can also be queried over the Internet [5].
A ThoughtTreasure server can be accessed using the Java-based client API [6] or by directly using the ThoughtTreasure server protocol [7].
ThoughtTreasure has been used to build various applications such as a DJ's assistant, a movie review question answering program, and a smart calendar.
ThoughtTreasure was begun by Erik T. Mueller in December 1993. The first version was released on April 28, 1996. Mueller established the company Signiform in 1997 to pursue commercial applications of ThoughtTreasure. However, the company was unsuccessful and Signiform closed its doors in 2000. Although the company's site is still being hosted, no further development of ThoughtTreasure has been released since then.