tough nuts, with primitives
schubert@cs.rochester.edu
Date: Fri, 8 Jun 90 17:41:22 -0400
From: schubert@cs.rochester.edu
Message-id: <9006082141.AA01533@ash.cs.rochester.edu>
To: interlingua@venera.isi.edu
Subject: tough nuts, with primitives
Cc: kr-standards@vaxa.isi.edu
MRG has - reasonably - asked what primitives I wanted to assume in
the various "tough nut" examples; e.g., "ABC_CORP is likely to fold
next month" is a whole lot easier to represent if we allow ourselves
a predicate is-likely-to-fold-next-month(x) than if we require
individual words to be somehow logically represented.
In general, I pretty much had in mind "fine-grained" translations,
i.e., essentially at the word level. After all, expressions like
"is likely to fold next month" mean what they do in virtue of the
meanings of the individual words and the way these happen to be put
together syntactically here (not to mention contextual effects).
So using hyphenation freely is in general a cheat -- it fails to do
justice to the way word meanings contribute SYSTEMATICALLY to the
meaning of larger constructs. For example, there would be no
analyzable connection between "is-likely-to-fold-next-month",
"is-unlikely-to-fold-next-month", "is-likely-to-make-record-profits-
next-year", etc. (Also, this would lead to explosive growth of the
primitive vocabulary and axiomatization.)
On the other hand, the ten sets of examples are aimed at specific
issues, and we can certainly gloss over the details of those
phrases whose internal structure is irrelevant to those issues.
So I am going to annotate the examples with what I would regard
as the allowable primitives, many of which will be hyphenated
phrases. Also, I am expecting the issue of indexicality (context-
dependence) to be finessed at this point by assuming, for instance,
that the reference of tense constructs, pronouns, and definite or
indexical NPs (like "next month") is known.
In providing primitives, I unfortunately can't go so far as to
say what the semantic TYPES of the primitives are. For instance,
I can't say whether "faulty" in (8a) should be a 1-place predicate
over individuals, a 2-place predicate with an individual argument
and a comparison-class argument, or a 1-place predicate modifier
(mapping predicates into predicates), or something else. That's
part of the problem! I do have views about these things, but that's
not the point -- I want the examples to "speak for themselves",
perhaps suggesting extensions to the logical types available in
the interlingua.
Of course, it would be nice to "dehyphenate" the primitives in a
later attempt to deal with the tough nuts; this may become feasible,
since many of the problems that the hyphens conceal are addressed
by one or another of the tough nuts.
TEN "TOUGH NUTS" FOR A GENERAL KR
---------------------------------
Note: Many of the examples that follow are fairly directly representable
in certain emerging episodic/situational/intensional/type-theoretic
representations geared towards natural language. If translation
from those representations to the interlingua requires complex
case-by-case circumlocutions, it is unlikely that KB's based on
them will be readily sharable via the interlingua.
Note: One slight change since the last version is that the definite
articles in (8c) have been changed to indefinites.
Note: Suggested primitives follow each example, in square brackets.
Presumed names start with a capital letter, while quantifiers,
predicates and functions (of whatever type) are in lower case.
In many cases, these may have to be augmented with additional
primitives not explicitly realized in the surface text; for
instance, we probably need "Now" (the time of assertion),
a "duration" predicate of some sort, predicates for specifying
ordering relations among events, implicit quantifiers, implicit
event (and other) variables, a set-forming operator, etc.
1. EVENTS, TIMES AND CAUSES
a. Company ABC has been manufacturing product X for three months.
[ABC, be-manufacturing-product, X, three, month]
b. Every company in region XYZ laid off some workers in May 1990,
causing the level of unemployment in XYZ to rise by 3%.
[every, company-in-region-XYZ, lay-off-some-workers, during,
May-1990, cause, level-of-unemployment-in-region-XYZ, rise-by-3%]
c. Company ABC sold no supercomputers in May 1990, causing a drop
in the value of its shares.
[ABC, sell, no (or not), supercomputer, during, May-1990, cause,
decrease, value-of-shares-of]
2. UNCERTAINTY
a. Company XYZ is likely to fold next month (July 1990).
[XYZ, likely, fold (or go-out-of-business), July-1990]
b. If company ABC has no sales this month (June 1990), it will
probably fold next month.
[if-then, (or implies), ABC, has-no-sales, June-1990, probably
(or probable), fold, July-1990]
3. UNRELIABLE GENERALIZATIONS
a. When a company releases a new product, it usually advertises it.
[when, usually (or usually-when, or usually-implies, etc.),
company, releases, new-product, advertises]
b. Computer programs with more than 2000 lines of code almost
always contain bugs.
[computer-program, contain-more-than-2000-lines-of-code,
almost-always (or almost-always-when, or almost-always-implies,
etc.), contain-bugs]
4. GENERALIZED QUANTIFIERS
a. Most/many/few of the subsidiaries of ABC_CORP are located in
California.
[most, many, few, subsidiary, ABC_CORP, located-in, California]
5. KINDS
a. A dodo is a kind of flightless bird. This kind of bird was
sometimes sighted in Mauritius, but is now extinct.
[dodo, kind (or kind-of), flightless, bird, some, time,
be-sighted-in-Mauritius, and, Now, extinct]
b. Some makes of automobiles with rotary engines are popular.
[some, make (or make-of), automobile-with-rotary-engine,
popular]
c. The cloth on this sofa is no longer being manufactured.
[the, (kind-of?), cloth, covering-this-sofa, not, still-being-
manufactured]
6. SETS
a. Smith, Green, and 18 other individuals pooled their assets
and formed a company.
[Smith, Green, 18, not-equal, individual, pool-assets-of,
form-company]
b. The "nuclear club" (the set of nations possessing nuclear
weapons) has at least 6 and at most 10 members. The known
members are the USA, the USSR, France, (etc.)
[Nuclear-club, equal, set (or set-of), nation-possessing-
nuclear-weapons, (number-of? cardinality-of?), >=, <=, 6, 10,
member-of, (known-that? known-to-be?), USA, USSR, France, etc.]
c. The actual number of people carrying the HIV virus is at least
twice the number of known carriers.
[number-of (or cardinality-of), person-carrying-the-HIV-virus,
>=, times, 2, (known-that? known-to-be?)]
7. NOMINALIZATION
a. Concealing profits is illegal.
[(nominalization operator? others?), conceal-profits, illegal]
b. To conceal profits is illegal.
[(nominalization operator? others?), conceal-profits, illegal]
c. For a company to conceal profits is illegal.
[(nominalization operator? others?), company, conceal-profits,
illegal]
8. MODIFIERS
a. User interface X degrades gracefully on mildly faulty inputs.
[X, degrade, graceful(ly), mild(ly), faulty, input-to]
b. The interlingua is a highly expressive knowledge representation.
[Interlingua, high(ly), expressive, knowledge-representation]
c. A new highway runs through the ABC mountains, alongside a much
older railroad track.
[new, highway, run (or has-path, etc.), through (or directed-
through, etc.), ABC-mountains, alongside (or ...), much, older-
than (or age, much-greater-than, ...), railroad-track]
9. MODALS
a. Company ABC suspects that a competitor obtained a copy of its
(ABC's) unreleased product X.
[ABC, suspect, that (or suspect-that), competitor-of, obtain,
copy-of, X]
b. Company ABC intends to force company XYZ to sell its assets.
[ABC, intend (or intend-that, or intend-to), force (or force-to),
sell-its-assets]
c. If company ABC had not sold its subsidiary XYZ, it would have
gone bankrupt.
[counterfactually-implies (or ...), ABC, not, sell, XYZ,
go-bankrupt]
10. OTHER FORMS OF INTENSIONALITY
a. Company ABC is seeking new customers.
[ABC, seek, new, customer]
b. Company ABC is designing a new supercomputer.
[ABC, design, new-supercomputer]
c. The new memory chip X resembles a 128-level Mayan pyramid.
[X, resemble, 128-level-Mayan-pyramid (a predicate which applies
to no actual objects)]