annotation of membership about amr-guidelines

[nschneid]

My understanding is that include-91 is restricted to sets where the collection and collected are alike in kind (when discussing numeric quantities), and so is not used for the relation between a member and the organization they belong to. See issue #3.

Some cases we have struggled with:

1. EMPLOYEE ~ COMPANY: currently :employed-by
2. EMPLOYEE ~ BOSS: not sure...is this covered by :employed-by?
3. UNIT ~ COMPANY: :part-of? (I think other projects use the term "subsidiary")
4. PERSON-MEMBER ~ ASSOCIATION: e.g., member of the historical society (not an employee)
5. INSTITUTIONAL-MEMBER ~ ASSOCIATION: NATO members
6. PERSON ~ OFFICIAL-ROLE-NAME: president of the company, director of human resources for the company. Might be captured well be noun frames.

I think cases (1), (4), and (5), at least, seem like quite similar concepts, but the term "employed by" suggests (1) and (2). Should we generalize to a :member-of role? (Alas, we do not usually use the word "member" for employees, but I can't think of a more general term....)

[biancab]

The suggestion was to use include-01, not include-91, since membership is an entity-group relationship, rather than a subset-superset relationship.

[nschneid]

I think it's a frequent enough case that it deserves a special AMR name; include-01 might be the appropriate reification. E.g., X :member-of Y could reify to include-01 :ARG1 X :ARG2 Y.

[nschneid]

Here's a proposal, though it would have to be modified if we start using noun frames:

Define membership as a relation between person (or personified entity) and an institution/association that they have joined/entered into, establishing a semi-permanent and often official status. Membership implies being associated in a particular capacity, like vice president, acting director, or simply member.

Let membership be expressed with serve-01. We will say :ARG0 is the member, :ARG1 is the group/institution/employer, and :ARG2 is the capacity (though I'm not sure the PropBank definitions quite line up). X :member-of Y would map to serve-01 :ARG0 X :ARG1 Y.

Examples:

• Dr. Smith of Tech Industries: [Dr. Smith] :member-of [Tech Industries] => [Dr. Smith] :ARG0-of (s / serve-01 :ARG1 [Tech Industries])
• president of Tech Industries: person :ARG0-of (s /serve-01 :ARG1 [Tech Industries] :ARG2 president)
  • Because "president" names the role, we fill in :ARG0 with person. You could think of this as a person who "presidents" Tech Industries.
  • Issue: Should these relational nouns always be unpacked into "person who serves as __", or only when they take arguments? Should "President Obama" be "president named Obama" or "person named Obama who serves as president"?
• Tech Industries' exuberant president: person :mod exuberant :ARG2-of (s / serve-01 :ARG1 [Tech Industries]) (i.e., an exuberant person who serves as president, not someone who fills the role of exuberant president)
• Tech Industries' vice president: person :ARG0-of (s / serve-01 :ARG1 [Tech Industries] :ARG2 [president :mod vice]) (i.e., someone who serves as vice president)
• Tech Industries' new president: person :ARG0-of (s / serve-01 :mod new :ARG1 [Tech Industries] :ARG2 president) (i.e., someone who serves newly as president)
• Dr. Smith's students: person :ARG0-of (s / serve-01 :ARG1 [Dr. Smith] :ARG2 student) (whether they are employed as his students or just taking his class, "student" describes their relationship)
• NATO member nations: nation :ARG0-of (s / serve-01 :ARG1 [NATO] :ARG2 member)
• U.S. president: person :ARG0-of (s / serve-01 :ARG1 [U.S.] :ARG2 president)

For divisions/units of companies, I am more inclined to think of it as a subpart relation:

• vice president of human resources at Tech Industries: one reading is "vice president of the HR department", the other is "vice president of the company whose role is directing human resources"

(p / person
   :ARG0-of (s / serve-01 
               :ARG2 (p2 / president :mod vice)
               :ARG1 (r / resource
                        :mod human
                        :part-of [Tech Industries])))

(p / person
   :ARG0-of (s / serve-01 
               :ARG2 (p2 / president :mod vice
                         :mod (r / resource
                                 :mod human))
               :ARG1 [Tech Industries]))

"One of the people over there" or "members of the over-80 crowd" [i.e., all the people over 80 years old] would not qualify as membership: instead, :subset-of/include-91 would be used for the former and include-01 for the latter.

[cbonial]

I think this issue definitely needs to be revisited. During training, it became obvious that the motivation behind selecting "employed-by" vs. "poss" is not clear to me. Does this depend on whether or not we know that the person is directly PAID by the company (as opposed to receiving stock dividends or some other compensation)? Aren't vice presidents, chairmen, ceos, like spokeswomen, "employed by" the company? How are we to be making this distinction?

[uhermjakob]

I agree that this needs to be improved. employed-by does not really capture the core of relations such as vice president etc. In many organizations, in particular non-profits, leadership positions are often pro bono. And CEOs of one company might serve on the board of directors of other companies, getting come compensation, but without being an employee of several companies.

Nathan made a proposal above using serve-01. I like the core of the proposal, but I think that serve-01 does not quite capture the breadth of cases, e.g. shareholders of a company; so I'm leaning towards a new frame such as:

(p / has-role-91
   :arg0 [role/office/function holder]
   :arg1 [organization]
   :arg2 [role/office/function])

[kevincrawfordknight]

i'm kind of liking has-role-91.

would be good to have examples & boundaries. how about john's mother? (probably not, because the family is not an organization?)

or better, examples of things to re-write (and not re-write) from wsj-100.

[nschneid]

Again, if we adopt noun frames this might all become a moot point. But one possibility is to distinguish has-role-91, which would apply to organizations and membership, from has-relation-91, which would apply to relations between two people (including kinship, friendship, employer-employee, consultant-client, doctor-patient, student-teacher, and the like). The arguments would be analogous.

We could try to go really deep and canonicalize the asymmetric relations, so "student of X" would become "person that X teaches", while "doctor" and "patient" would evoke treat-03, etc. We have a small precedent with "pianist" being handled like "piano player," which is a natural English paraphrase. But trying to do this in general seems like a can of worms—especially where the relation terms encode different amounts of information for different participants. Consider gendered familial relations: we can infer from "X's son" that X is a parent, and from "Y's mother" that Y is a child, but the genders of X and Y are unspecified. So converting "X's son" to "person that X parents" would remove information.

[ArchnaBhatia]

I also like Nathan's proposal above if we replace "serve-01" with the relation "has-role-91", we use it for relationships between organizations or for relationships between organizations and people. The analogous "has-relation-91" would be good for relationships between people, but since there is a concern about losing the gender information, how about the following?

We have a slightly different architecture for "has-relation-91": it has arg0 and arg1 which are both relations' names for which further information is provided as available, see below:

(1) For "X's son" the AMR would be:

(h / has-relation-91
      :ARG0 (s / son)
      :ARG1 (p / parent :name (n / name :op1 "X")))

This AMR is also good for "a person that X parents" and we don't lose any gender information.

For "X's son Bill", the AMR would be:

(h / has-relation-91
      :ARG0 (s / son :name (n2 / name :op1 "Bill"))
      :ARG1 (p / parent :name (n / name :op1 "X")))

[uhermjakob]

At the AMR workshop July 16-18, 2013, participants (Kevin, Martha, Claire, Kira, Afton, Natalie and Ulf) decided to introdude a new general frame for having a role in an organization with the understanding that for some select cases (such as professor) we might use a more detailed special frame. This new frame also covers membership relationships.

(h / have-org-role-91
   :arg0 [role/office/function holder]
   :arg1 [organization]
   :arg2 [role/office/function])

Examples

President Obama

(h / have-org-role-91
  :ARG0 (p / person :name (n / name :op1 "Obama"))
  :ARG2 (p2 / president))

United Nations Secretary General Ban Ki-moon traveled to Egypt.

(t / travel-01
  :ARG0 (p / person :name (n / name :op1 "Ban" :op2 "Ki-moon")
        :ARG0-of (h / have-org-role-91
              :ARG1 (o / organization :name (n3 / name :op1 "United" :op2 "Nations"))
              :ARG2 (s / secretary
                    :mod (g / general))))
  :ARG1 (c / country :name (n2 / name :op1 "Egypt")))

Spain is a member of NATO.

(h / have-org-role-91
  :ARG0 (c / country :name (n / name :op1 "Spain"))
  :ARG1 (m / military :name (n2 / name :op1 "NATO"))
  :ARG2 (m2 / member))

[mgeorgescu]

Hi all,

Should this new role be used from now on? Please note that we have not used it for the July data release.

Should this role be used ONLY when we are dealing with a person occupying a role in an organization and the role and organization are specified in the sentence?

In the case of a person having multiple roles like in the sentence below would the AMR below be a good representation?

Nancy Scheper - Hughes is Founding Director of organs Watch , a professor , and the Director of University of California at Berkeley 's medical anthropology program .

Or should "professor" be :arg2 in a different (h / have-org-role-91) ? What about the "director" of a program? Given the fact that the program is not an organization, the (h / have-org-role-91) should not be used?

Another example:
20080710 director of the London - based International Institute for Strategic Studies 's non - proliferation program and former US State Department official Mark Fitzpatrick stated that an Iranian photograph showing a cluster of missile launches from a desert range was altered to add a 4 th missile .

Thank you.

[ArchnaBhatia]

If it were "a professor at University of California", in that case using (h / have-org-role-91) would be better, but as it is in the given sentence, I think this is a good representation.

Also for the director of the anthropology program, I think this is a good representation- here the person (PB agent ARG0) directs the program (PB theme ARG1).

[uhermjakob]

We should use have-org-role-91 for director under :op3 as well.
Let's remember not to verbalize director under :op3, just as under :op1.
I think that the "org" in have-org-role-91 should be interpreted generously enough to accommodate divisions of organizations, projects etc. (but not generic sets of people such as cancer survivors).

Even for professor I'm leaning towards using have-org-role-91. One can be a doctor without working for a specific organization, but if I'm not mistaken, you can only be a professor when you have the proper position at a proper university. I don't think that there are any freelance professors, just as there are no freelance presidents.

[biancab]

So have-org-role-91 should be used instead of :employed-by in such cases as <"We have no useful information on whether users are at risk ,'' said James A. Talcott of Boston 's Dana - Farber Cancer Institute">? The current consensus representation for this is:

(s / say-01
:ARG0 (p / person :name (j / name :op1 "James" :op2 "A." :op3 "Talcott")
:employed-by (r / research-institute :name (d / name :op1 "Dana-Farber" :op2 "Cancer" :op3 "Institute")
:location (c / city :name (b / name :op1 "Boston"))))

[nschneid]

So have-org-role-91 should be used instead of :employed-by in such cases as <"We have no useful information on whether users are at risk ,'' said James A. Talcott of Boston 's Dana - Farber Cancer Institute">?

Should X :employed-by Y reify to X :ARG0-of (h / has-org-role-91 :ARG1 Y)? Or is has-org-role-91 only intended for cases where the role has a particular job title more specific than "employee"?

[nschneid]

Even for professor I'm leaning towards using have-org-role-91. One can be a doctor without working for a specific organization, but if I'm not mistaken, you can only be a professor when you have the proper position at a proper university. I don't think that there are any freelance professors, just as there are no freelance presidents.

One could be a professor by profession. :) Presumably it requires having been employed by at least one university, just as a judge is presumably employed by a court, or a patent inspector presumably works for a patent office. Do we really want to decide for each role concept whether its lexical semantics implies an organization? It seems like it could get rather tricky...one could act as a teacher without working for an organization, but what about student teacher or head teacher or gym teacher? Is a chef someone who has certain training/qualifications, or someone employed as a cook at a restaurant? etc.