An Intelligent Analyzer and Understander of English

Yorick Wilks

1975, ACM

2

Introduction

Semantics– Natural Language Processing cannot do this directly

– Minsky argued that there could be no machine translation without a system that, in an adequate sense, understood what it was trying to translate.

Preference Semantics– Contains no conventional grammar for analysis or gene

ration, its task is performed by a strong semantics.

3

A System of Semantics Based Language Analysis(1)

Semantic Block– A representation of a fragment text.

It consists of templates bound together by paraplates and common sense inferences.

These items consist of formulas which in terms consist of elements.

4

A System of Semantics Based Language Analysis(2)

Semantic items represent text items as follows:

Item in semantic representation

Corresponding text items

formula English word sense

template English clause or simple sentence

semantic block English paragraph or text

5

Semantic Elements(1)

Semantic elements are 70 primitive semantic units used to express the semantic entities, states, qualities, and actions about which humans speak and write.

6

Semantic Elements(2)

The elements fall into five classesa) Entities: MAN (human being), STUFF (substances),

FOLK (human groups), etc. b) Actions: FORCE (compels), CAUSE (causes to happ

en), FLOW (moving as liquids do), etc. c) Type indicators: KIND (being a quality), HOW (bein

g a type of action), etc.d) Sorts: CONT (being a container), GOOD (being mora

lly acceptable), THRU(being an aperture), etc.e) Cases: TO (direction), GOAL (goal or end), SUBJ (ac

tor or agent), etc.

7

Semantic Formulas(1)

Semantic formulas are constructed from elements. These elements express the senses of English words; one f

ormula to each sense. The formulas are binarily bracketed lists of whatever depth

is necessary to express the word sense. The most important element is the head of the formula, al

ways at their rightmost. It expresses the most general category under which the word sense in question falls.

An element that is used as a head can function within formulas as well. Formulas can be thought of and written out as binary trees of semantic primitives.

8

Semantic Formulas(2)

9

Semantic Templates(1)

Semantic elements have been explained by seeing how they functioned within formulas. One level higher of formulas, the third kind of semantic item in the system, are to be explained by describing how they function within templates.

A template consists of a network of whole formulas, and its connectivity is between an agent- , action- , and object-formula.

10

Semantic Templates(2)

Giving the example that, "Small men sometimes father big sons", these formulas will contain two sequences of head elements, illustrated as follows:

11

Templates and linguistic Syntax

Two bared templates are matched as illustrated as follows, but it has no reason so far as to prefer one to other.

12

Case Ambiguity(1)

Templates are matched with each fragment of the text.

There are then complex routines for establishing contextual ties between these templates separated by fragmentation.

It is both psychologically and computational important for dealing with text containing realistically long and complicated sentences.

13

Case Ambiguity(2)

How does the system resolve ambiguity? – The input text is fragmented initially in the actual

implementation of the system. Templates are matched with each fragment of the text.

– TIE routines attach the structures for separated fragments back together after the preliminary assignment of template structures to individual fragments.

– However, it is important to note that a preference is between alternatives. If the only structure derivable does not satisfy a declared preference, it is then accepted anyway.

14

Paraplates and Case Ambiguity

The paraplates are essentially patterns that span two templates, which named by the writer as mark and case templates, where the mark template generally precedes the case template.

If the predicates are all satisfied by the contents of the two templates, then that paraplate is considered to match onto the two templates and the case ambiguity of the preposition that functions as the pseudo-action in the second template is solved.

15

Anaphora and References

When does common sense inference rules apply? – Common sense inferences are only called in wh

en TIE is unable to resolve outstanding anaphoras. A process of extraction is then done, the common sense rules subsequently apply to these extractions and the relevant templates as the procedure attempts to build up a chain of extractions and inferences.

16

The Generation System for French

How does the generation system for French work? – The system of generation patterns called stereotypes is

required for translating into French. The sense-pairs and formulas are carried to the block when a word sense is selected during analysis.

– Complex stereotypes are strings of French words and functions, that is the interlingual context of the sense-pair.