2 Five Generations of MT 13
means always, the human reader will have no difficulty in resolving the ambiguity
through utilisation of his background knowledge, no counterpart of which could
possibly stand at the disposal of computers” (1962).
The immediate historical context of Bar-Hillel’s argument was the performance
of early syntax analysers, which, according to legend, were capable of producing
upwards of 10 grammatical parsings of sentences like “Time flies like an arrow”.
With respect to standard dictionary information, any of the first three words in the
sentence could be taken as a possible verb. To see “time” as the verb, think of the
sentence as command with the accent on the first word; to see “like” as the verb,
think of the sentence as expressing the tastes of a certain kind of fly, and so on.
The standard reaction to such syntactic results was to argue only showed the
need for linguistic semantics, so as to reduce the “readings” in such cases to
the appropriate one. Bar-Hillel’s response was to argue that it was not a matter
of semantic additions at all, but of the, for him, unformalisable world of human
knowledge.
The contrast can be seen by looking at our everyday understanding of so simple
a sentence as “He paddled down the river in a canoe”. The standard machine parser,
working only with grammatical information, will not be able to decide whether the
clause “in a canoe” attaches to “paddled” or “river”. The first reading, the correct
one of course, tells you how he went down the river. The second implies that we
went down a river that happened to be inside a canoe – the same structure that
would be appropriate for “He paddled down the river in an unexplored province of
Brazil”. The purely syntactic parser has no way of distinguishing these two possible
“readings” of the sentence and, furthermore, there is a difference of opinion as to
how the information that would resolve the problem should be described. Those
who take a more “linguistic semantics” view would say that it is part of the meaning
of “canoe” that those objects go in rivers and not vice versa; whereas those of an Al
persuasion would be more likely to say that it is merely a fact about our world that
canoes are in rivers. At bottom, there is probably no clear philosophical distinction
between these views, but they do lead to different practical results when attempts
are made to formalise and program such information.
Bar-Hillel went further and produced an example (the best-known in the history
of MT) proving, as he thought, the impossibility of MT. In brief, Bar-Hillel’s exam-
ple was the following children’s story:
Little John was looking for his toy box.
Finally he found it.
The box was in the pen.
John was very happy.
Bar-Hillel’s focus is on the third senteiipnce, “the box was in the pen”, whose
last word we naturally interpret in context as meaning playpen and not writing
pen. Bar-Hillel argued persuasively that to resolve this correctly requires knowl-
edge of the real world, in some clear sense: at least in the sense that the diffi-
culty cannot be overcome in terms of some simpleminded “overlap of concepts”,