Robotics Reference
In-Depth Information
From these ten sentences, which are made up from a vocabulary of only
38 words, Yngve derived 77 rules, from which it would be possible to
generate all ten of the sentences and a huge number of other sentences
as well, by randomly choosing nouns to fill the NOUN slots in his rules,
randomly choosing verbs to fill the VERB slots, and so on. One type of
sentence that can be produced by this grammar is
Theengineisblack,oiled,...andshiny.
where the“...” couldsensiblybe leftblankorfilledinbyoneormore
of the other adjectives in the vocabulary: little, polished, big. Yngve
calculated that, even with a tight constraint on the number of adjectives
that could be used in succession in this way, there were more than 10 20
different sentences that could be generated from the 38-word vocabulary,
every one of them conforming to the 77 rules of his simple grammar.
The randomness in Yngve's approach was not suitable for Sheldon
Klein, whose work at the University of Wisconsin on novel and story
writing in the late 1960s required a sentence generator that could pro-
duce more meaningful text. Klein's program was based on a type of gram-
mar called a dependency grammar in which each rule ensured that most of
the words in a phrase were associated with (i.e., depended on) the head of
the phrase (the most important word). For example, in “The man rides a
bicycle” the word “bicycle” depends on “rides”, while in the sentence “A
bicycle is a vehicle with wheels” the word “vehicle” depends on “is”, and
the word “is” depends on “bicycle”. So it would be acceptable, within a
grammar containing these dependencies, to generate “The man rides a
vehicle with wheels”, which certainly represents an improvement on the
output typical of Yngve's program.
But after Klein's work on using dependency grammars, rather little
progress was made in this field over the subsequent 30 years, and it was
not until the late 1990s that text generation programs were able to pro-
duce passages of well-structured, meaningful text extending to several
paragraphs.
TALE-SPIN
The earliest attempt at computerized story-writing was James Meehan's
research for his PhD thesis at Yale University, published in 1976. Mee-
han's program, TALE-SPIN, was the result of his attempts to investigate
Search WWH ::




Custom Search