Closed vcvpaiva closed 6 years ago
I'm afraid it's not so easy. While tempting just to map to Sand, this adjective is much broader, refer to things growing in sand, or of a sandy consistency. 'part' is an incorrect mapping, I agree. We don't have a good solution for this. One possibility might be to create something like (SimilarToFn ?X).
@apease like when we are talking about the demonyms, adjectives don't have an obvious mapping.
Demonyms are not adjectives, they are nouns and have an obvious meaning, "inhabitant or native from a given place X".
"sandy" (like all pertainyms) pertains or is related to "sand", but the relation can be very far from concrete. so a "sandy beach" is really a beach full of sand. when "sandy" refers only to the colour, like "sandy hair" it's a different synset (00244199-a ) but when something grows in sand, they also get to be "sandy" as in "sandy grasses". I still think it's perfectly fine to map it to "pertaining to Sand", (other mappings are like so) but if @apease doesn't want it, this is OK by me.
just to make sure in SICK we have pertainyms: snowy, sandy, grassy, rocky, furry, elderly, shiny, silky, foamy, hungry, lovely, hairy, healthy, in this case.
The challenge is that Sand in SUMO doesn't mean "pertaining to sand". It means an instance of the Substance Sand. In the same vein Brazillian can mean a person from Brazil or "pertaining to Brazil" like a Brazillian steak (I'm hungry now). So, one possibility for what we would need to do would be to create (PertainingToFn ?X) and then reify all the cases that are lexicalized in WordNet, such as
(instance Brazillian-PertainingTo RelationalAttribute) (=> (attribute ?X Brazillian-PertainingTo) (attribute ?X (PertainingToFn Brazil)))
and then map
02966829 01 a 01 Brazilian 0 002 + 08853741 n 0101 \ 08853741 n 0101 | of or relating to or characteristic of Brazil or the people of Brazil &%Brazilian-PertainingTo=
On 03/03/2017 05:09 PM, Valeria de Paiva wrote:
Demonyms are not adjectives, they are nouns and have an obvious meaning, "inhabitant or native from a given place X".
"sandy" (like all pertainyms) pertains or is related to "sand", but the relation can be very far from concrete. so a "sandy beach" is really a beach full of sand. when "sandy" refers only to the colour, like "sandy hair" it's a different synset (00244199-a ) but when something grows in sand, they also get to be "sandy" as in "sandy grasses". I still think it's perfectly fine to map it to "pertaining to Sand", (other mappings are like so) but if @apease https://github.com/apease doesn't want it, this is OK by me.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ontologyportal/sumo/issues/61#issuecomment-284112792, or mute the thread https://github.com/notifications/unsubscribe-auth/ACBvHTlbN5hgi2pUPcWFnzBnRFV62HXWks5riLm_gaJpZM4MSYWS.
Adam Pease http://www.ontologyportal.org http://www.adampease.org
Yes, this is what I have been advocating since our paper about the gentilics. (Gentle with the Gentilics. ACL Anthology, 23 May 2016. ) the thought is that for locations like countries this is not so full of nuances. the second thought is that we need to do demonyms too (the noun "Brazilian")
OK, according to our terminology in Gentle with the Gentilics paper, I was talking about the pertainyms, my mistake.
The @apease's suggestion of creating PertainingToFn
is not very intuitive for me. What is the meaning of the term (PertainingToFn Brazil)
? But all those abstract definitions are hard to understand anyway. SUMO has:
Quality
and Attribute
as subclasses of Abstract
.attribute
, property
where attribute
is subrelation of property
.and they are defined as:
(attribute ?OBJECT ?PROPERTY)
means that ?PROPERTY is a Attribute of ?OBJECT.property
is defined as "this Predicate holds between an instance of Entity and an instance of Attribute"Some examples:
(=> (equal ?ATTR1 ?ATTR2)
(forall (?THING)
(<=> (property ?THING ?ATTR1)
(property ?THING ?ATTR2))))
(<=> (attribute ?INDIVIDUAL Buddhist)
(member ?INDIVIDUAL Buddhism))
(<=> (attribute ?I
(BeliefGroupMemberFn ?BG))
(member ?I ?BG))
What is not clear to me is the difference between the predicates attribute
and property
.
With respect to the wordnet mappings. I belive these reifications will inflate the ontology with needless names. This reminds me why people started to use lambda calculus for NL semantics (from the book [1] that I am reading). Page 39:
As in FOL, HOL allows us to provide terms for the lexical items in an expression like Sandy dazzled Chris. Sandy and Chris are represented by individual constants, which are interpreted as individuals, and dazzled is represented by a two-place relation symbol, which is interpred as binary relation. but unlike FOL, [using \lambda calculus] we can in addition provide a term corresponding to the meaning of a verb phrase dazzled Chris. The closest approximation in FOL would be dazzle(x, Chris), but this is a formula with a free variable, which is not at all the same kind of expression as run, a unary predicate symbol, which would be the translation of the verb phrase run. The \lambda-calculus provides a language in which we will be able to assign all of our constituents first-class logical expressions.
This is precisely what is happing here, the mappings could be much more consistent with an intermediary layer of lambda terms.
[1] B. Carpenter, Type-Logical Semantics. MIT Press, 1998, pp. 1–587.
What is the meaning of the term (PertainingToFn Brazil)?
of or relating to or characteristic of Brazil or the people of Brazil
I believe we have 6177 adjectives in this category in PWN (3547 without Portuguese translations), admittedly most of them (cacodylic, palingenetic, myrmecophytic) I don't care for. but I do care for the ones that are likely to show up in the news, in Wikipedia, in daily conversations, in the DHBB.
so I do not think
With respect to the wordnet mappings. I belive these reifications will inflate the ontology with needless names.
and while I find it nice to hear you're reading old friends, Bob's argument about using HOL for the predicate "dazzle" in type-logical semantics has nothing to do in this discussion, IMHO.
@vcvpaiva you missed my point entirely. Anyway...
@apease maybe you do not want to do all pertainyms, but making sure you have all the gentilics (pertainyms related to locations) would be great already. this was what @livyreal wanted to do for the gentilics paper.
@arademaker I agree with you on
What is not clear to me is the difference between the predicates attribute and property.
but yes, missing your points entirely is my life's work.
There are a couple of opposing considerations I have on this topic. On the one hand, I'd like to keep the SUMO-WN mapping formalism simple. That makes it easy to use and adopt. I'd also like to keep SUMO as "clean" as possible - terms that can be easily stated as formulas shouldn't be created unless that new term is likely to have wide use in other axioms. Functional terms are a good way to avoid the proliferation of terms that differ only in one simple way, like GovernmentFn instead of having GovernmentOfUS, GovernmentOfBrazil etc. But pertainyms like American-like, Brazil-like etc are unlikely to be reused extensively and directly. One way to handle this general issue is to put more complex relationships between logic and language into the Semantic Rewriting rule set, which could be viewed as similar to lambda expressions. But the complexity and performance of that rule set would rise rapidly and make Sigma's NLP harder if I put things of this sort into the rule set. With all that said, I think the best solution is for me just to add the gentilics as terms to SUMO, so we can keep the SUMO-WN mapping simple.
On 03/04/2017 06:49 AM, Valeria de Paiva wrote:
@apease https://github.com/apease maybe you do not want to do all pertainyms, but making sure you have all the gentilics (pertainyms related to locations) would be great already. this was what @livyreal https://github.com/livyreal wanted to do for the gentilics paper.
@arademaker https://github.com/arademaker I agree with you on
What is not clear to me is the difference between the predicates attribute and property.
but yes, missing your points entirely is my life's work.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ontologyportal/sumo/issues/61#issuecomment-284155977, or mute the thread https://github.com/notifications/unsubscribe-auth/ACBvHWWfXhpwTqNWo6ONOtJX6xzC33HGks5riXoCgaJpZM4MSYWS.
Adam Pease http://www.ontologyportal.org http://www.adampease.org
hi @apease, do you feel like doing the gentilics as you said above?
With all that said, I think the best solution is for me just to add the gentilics as terms to SUMO, so we can keep the SUMO-WN mapping simple.
I think Livy and I would like to help, but would appreciate a leading example from you!
Hi Valeria, I think we'd create SUMO terms corresponding to hyponyms of http://sigma.ontologyportal.org:8080/sigma/WordNet.jsp?synset=109620078&flang=SUO-KIF&lang=EnglishLanguage&kb=SUMO
First, I'll create a couple of terms to simplify each gentillic
(subclass NationalPerson GroupOfPeople)
(instance GentilicFn UnaryFunction) (domain GentilicFn 1 GeographicArea) (rangeSubclass GentilicFn GroupOfPeople) (documentation GentilicFn EnglishLanguage "A &%GroupOfPeople defined by their &%birthplace or where they &%Residence they &%inhabit is &%located.")
(=> (instance ?P (GentillicFn ?N)) (or (exists (?R) (and (located ?R ?N) (inhabits ?P ?R))) (birthplace ?P ?N)))
Then you could just create a relatively tedious file
(equal NewZealander (GenilicFn NewZealand)) (equal NewYorker (GenilicFn NewYork)) (equal Brazillian (GenilicFn Brazil)) ; etc
Then you'd modify the file https://github.com/ontologyportal/sumo/blob/master/WordNetMappings/WordNetMappings30-noun.txt
09694894 18 n 01 Brazilian 0 003 @ 09731571 n 0000 #m 08853741 n 0000 ~ 09695019 n 0000 | a native or inhabitant of Brazil &%EthnicGroup@
to be
09694894 18 n 01 Brazilian 0 003 @ 09731571 n 0000 #m 08853741 n 0000 ~ 09695019 n 0000 | a native or inhabitant of Brazil &%Brazilian=
and similarly for the NewYorker, NewZealander etc.
all the best, Adam
On 01/29/2018 02:08 PM, Valeria de Paiva wrote:
hi @apease https://github.com/apease, do you feel like doing the gentilics as you said above?
With all that said, I think the best solution is for me just to add the gentilics as terms to SUMO, so we can keep the SUMO-WN mapping simple.
I think Livy and I would like to help, but would appreciate a leading example from you!
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ontologyportal/sumo/issues/61#issuecomment-361403649, or mute the thread https://github.com/notifications/unsubscribe-auth/ACBvHa0RpZZVAwiu55b4Zs9AUgm15OK5ks5tPkFMgaJpZM4MSYWS.
Adam Pease http://www.ontologyportal.org http://www.adampease.org @apease_ontology on Twitter
ok @apease we will create the tedious file that starts with
(equal NewZealander (DemonymFn NewZealand)) (equal NewYorker (DemonymFn NewYork)) (equal Brazillian (DemonymFn Brazil)) ... and send it to you by email.
closing this issue, for the time being.
Reopening just to CORRECT that Adam meant "demonyms" not gentilics. Thus
We create a couple of terms to simplify each demonym:
(subclass NationalPerson GroupOfPeople) (instance DemonymFn UnaryFunction) (domain DemonymFn 1 GeographicArea) (rangeSubclass DemonymFn GroupOfPeople) (documentation DemonymFn EnglishLanguage "A &%GroupOfPeople defined by their &%birthplace or where the &%Residence they &%inhabit is &%located.")
(=> (instance ?P (DemonymFn ?N)) (or (exists (?R) (and (located ?R ?N) (inhabits ?P ?R))) (birthplace ?P ?N)))
Then we will create the tedious file that starts
(equal NewZealander (DemonymFn NewZealand)) (equal NewYorker (DemonymFn NewYork)) (equal Brazilian (DemonymFn Brazil)) ;
etc...
this is an easy one, sandy in
00142040-a (resembling or containing or abounding in sand; or growing in sandy areas; "arenaceous limestone"; "arenaceous grasses") arenaceous • sandy • sandlike
should map to
Sand (subsuming mapping)
(documentation Sand EnglishLanguage "Sand is loose fragments of minerals or rocks. Smaller than gravel and larger than silt and clay, sand particles range from 8/ 10,000 to 8/ 100 inch (0.02 to 2 millimeters) in diameter. Sand is formed by the Erosion of rocks through the action of water, ice, or air.") Geography.kif 4132-4135
instead of part, which means nothing...