Causal reconstruction is the task of reading a written causal description of a physical behavior forming an internal model of the described activity and demonstrating comprehension through question answering This task is difficult because written descriptions often do not specify exactly how referenced events together. This article characterizes the causal reconstruction problem presents a representation called transition space, which portrays events in terms of transitions or collections of changes expressible in everyday language and describes a program called PATHFINDER, which uses the transition space representation to perform causal reconstruction on simply English descriptions of physical activity PATHFINDER works by identifying partial matches between the representations of events and using these matches to form causal chains fill causal gaps and merge overlapping accounts of activity. By applying transformations to events prior to matching PATHFINDER is also able to handle a range of discontinuities arising from a writers use of analogy or abstraction.


Humans often learn about the causal workings of physical systems by reading written descriptions of the sort appearing in encyclopedias, reports and user manuals this article presents research in getting programs to read and reason on the basis of such descriptions [7, 8].

This task is both important and difficult It is important because written causal descriptions constitute not only an abundant resource for use in constructing intelligent user manuals design documentation systems and planning /diagnosis systems, but also a convenient medium for interaction with these systems during operation Moreover, these descriptions cover a wide range of phenomena humans find difficult to describe by other means: complex interactions such as combustion and phase changes, intuitive concepts such as sounds, paths and collections, and metaphorically-modeled activities such as radio signals “spreading” in space.

The task is difficult because written causal descriptions rely heavily on the reader’s ability to supply missing objects and events, identify relations between parts of the description and perform inference from the information provided. One especially difficult subtask is that of determining how the events in a description fit together into causal chains or overlapping accounts of activity as often these relationships are left implicit by the description.

 As an examples, consider the following excerpt taken from the opening paragraph of the entry for ”camera” in the Encyclopedia Americana [21].BSTRACT


Given this description, a human previously unfamiliar with the operation of a camera should be able to answer nontrivial questions such as the following.