amr-guidelines's Issues
Answers to yes/no questions
In sentences like the following:
- " Yes , that is true , "
- Yes , that 's it .
There were different opinions in the initial annotation on how to approach this:
- :mode affirmative --> which we don't have in our guidelines and probably if introduced would trigger the need to revisit all affirmative sentences (probably not desirable)
- yes dropped altogether and only the "that is true" annotated --> seems incomplete but may be acceptable.
For the post-edited little prince annotation, we went for the superficial approach:
(s / say-01
:ARG0 (p / prince
:mod (l / little))
:ARG1 (a / and
:op1 (y / yes)
:op2 (t / true
:domain (t2 / that))))
Should we go as far as introducing the concept "answer" (to make use of the argument structure) OR "confirm" in the case of sentence #2? and "infirm" in the case of a "no" answer (No, that is not true.)
Expressing mathematical calculations: equal-01
In sentences like:
Three and two make five .
We tried to better express the mathematical calculation by introducing the concept equal-01. Do you think this is acceptable? If yes, this should probably be added to the guidelines.
We drew on previous example from consensus:
[1] consensus isi_0002.148 (snt. 185 in workset guidelines, last updated on Tue Dec 11, 2012)
The size of Antarctica is as large as the United States and Mexico together.
(e / equal-01
:ARG1 (s / size
:poss (c / continent :name (n / name :op1 "Antarctica")))
:ARG2 (s2 / sum-of
:op1 (s3 / size
:poss (c2 / country :name (n2 / name :op1 "United" :op2 "States")))
:op2 (s4 / size
:poss (c3 / country :name (n3 / name :op1 "Mexico")))))
- In the case of using "equal-01" to express the mathematical calculation, the concept is missing from the sentence but it can easily be inferred, we would opt to add it:
Going from:
The sixth planet was ten times larger than the last one .
High frequency named entity: Kashmir
What NE type should Kashmir be? It's a region of a continent that's not administered by a particular country. Here's Wiki take:
Dogra rule, under the paramountcy (or tutelage) of the British Crown, lasted until 1947, when the former princely state became a disputed territory, now administered by three countries: India, Pakistan, and the People's Republic of China.
Should this be a "province" like Tibet?
In FS, I see this variously being called state, world-region, country-region, etc.
Role for ordinal numbers?
Ordinal numbers provide a distinctive type of modification and form a special morphosyntactic class in many languages. Do they deserve a special non-core role, e.g. :ord
(instead of :mod
)?
(p / planet
:ord 2)
the second planet
In contrast:
(h / hand
:mod (s / second))
the second hand (of a clock)
We would have to decide what to do about expressions that reference a sequence/ordering, e.g. second in a long line of kings or second to last of the finalists.
Annotating "Anti" and "pro" prefixes > oppose and favor?
In order to avoid superficial annotations like :prep-against for examples some of the examples below:
anti-virus program
anti-submarine aircraft
anti-missile system
anti-Syrian journalist
we introduced the concept "oppose-01"
and for "pro-Syrian regime", we opted for "favor" and "preferred annotation" for this initial release.
These 2 alternatives seemed better than prep-against or keeping the hyphenated concept.
Any feedback is appreciated.
Thank you.
Negative indefinite pronouns
Stick to the negative pronoun or map it to the positive one with negative polarity?
We tended to opt for the 2nd alternative.
nobody/ no one --> positive indefinite pronoun "anyone" OR "somebody" :polarity - ?
nothing --> positive indefinite pronoun "anything" :polarity - ?
none --> "one" OR "any" :polarity -
Could we try to clarify this in the guidelines as there has been some disagreement on this topic?
:mode exclamation
Problem
- ":mode exclamation" has been annotated very inconsistently (examples below).
Options
- drop :mode exclamation (favored by Ulf)
- expand in guidelines
Questions:
- Is ":mode exclamation" just a semantic glorification of ":punctuation !" ?
- What is the difference between the following two sentences?
-
This is great.
-
This is great!
-
- What about multiple !! ?
Consensus (?)
- ":mode exclamation" is not used in imperatives.
Examples of inconsistencies:
- Inconsistencies in consensus:
- wb-eng-0003.1 ("Braawwk !") without :mode exclamation
- wb-eng-0003.2 ("Braawwkkk !") with :mode exclamation [dropped for release]
- Annotators use all of these (and probably more):
-
:mode exclamation
-
:mode (e / exclamation)
-
:mode exclamatory
-
:mode (e / exclamatory)
-
Extra-sentential context
Case 1:
- Previous sentence: He said, "Yes.
- Current sentence: That's true.
- Should the second sentence, as an ongoing quote, be annotated with a say-01 at the top?
Case 2:
- Previous sentence: Wait?
- Current sentence: For what?
- Can we use the previous sentence, to annotate as (w / wait-01 :ARG2 (a / amr-unknown)) ?
Co-referential impossible to make due to sub-structure
In the example below, we would have to annotate twice the "Philippines" and would lose the co-reference:
U.S. military trainers are aiding officials in the Philippines in attempting to remove Islamic guerrillas from the southern Philippines .
A suggestion to preserve the co-reference would be:
remove-01
...
:arg2 location
:mod south
:location c
"location" would be the most generic category, but country-region can work just as well.
New roles for AMR 1.5
Let's propose new roles in this thread. These should come from real examples and be things we've seen more than once. I can start with
:size
Expressing frequency
Should we revisit the way we express frequency?
Will the 2 expressions "ever day" and "daily" get the same AMR and if yes, should it be the one above?
--> similar to the way we express time
- Should we be using temporal-quantity and unit/op and quant?
Expressing frequency in the case of "once" [1 time], twice [2 times] and "x times":
- Keep or drop "times"?
She cleans the house once a week. [She cleans the house weekly.]
They go on holiday twice a year.
To some extent "times" is a frequency indicator which can be "understood" from the ": frequency" role itself, therefore, dropping it altogether would result in a somewhat easier annotation.
They go to the gym three times a week.
Possibly more problematic : (every) once in a while, every so often
High frequency phrase: global X
We've noticed that "global" is being treated a bit inconsistently. It's generally either :mod globe or :mod global. It seems to me it should be stemmed to "globe."
It's used in phrases like:
Africa has 90 percent of the global diseases.
Sheni stated that Africa will find it difficult to influence global research priorities in its favor without a forum.
Sheni attended a four-day meeting to discuss global health at Arusha in northern Tanzania.
:subset (include-91), :example, :part/:subevent, :consist-of, and :employed-by
Proposed guideline:
- If there is an obvious verb frame that matches the statement of inclusion/containment, use it:
- supplies, including food, medicine, tarpaulins...:
include-01
- supplies, including food, medicine, tarpaulins...:
- If a category, set, or group is being illustrated with one or more members, use
:example
:- animals such as giraffes
- high tech companies like IBM and Google
- For sets of (more or less) homogeneous members under discussion—whether they are entities or events—use
:subset
:- the tallest girl on the team
- series of tremors
- If there is some established whose constitution is described, use
:consist-of
—this applies to physical substances as well as measure phrases:- ring of gold
- cloud of dust
- team of researchers
- If there is a whole/superstructure whose substructures are distinguished in kind, use
:part
or:subevent
:- house's roof
roof :part-of house
- my hands
hand :part-of i
- Newsweek , a unit of the Washington Post Co.
[Newsweek] :part-of [Washington Post Co.]
- Three Roman legions were annihilated in the Battle
annihilate-01 :subevent-of battle-01
- I pass the resort on my way to work.
pass-02 :subevent-of go-02
- house's roof
- Employers/employees are related with
:employed-by
:- Lorillard spokeswoman:
spokeswoman :employed-by [Lorrilard]
- (is the CEO example on this page out of date?)
- Lorillard spokeswoman:
Thoughts?
Location names within names of entities
When the location of a named entity is included in the name itself, do we treat this as a location, or as part of the name?
For example: Ecological organization Greenpeace stated on June 21, 2002 that Russia helped Iran to develop nuclear weapons by building Iran's Bushehr atomic power station.
Bushehr is the city in which the power station exists, and the rest isn't capitalized, so I initially assumed this should have an AMR where Bushehr is the :location of the power station.
However, subsequent mentions use "Bushehr" to stand in for the power plant itself:
Iran has said the plant is being built only for civilian energy purposes and allows regular inspections of Bushehr by the International Atomic Energy Agency.
Experts say Bushehr could in theory become operational as early as September 2003.
Should Bushehr therefore be treated as the name of the power station, and therefore be left semantically unanalyzed as the location? This also seems somewhat at odds with our guideline analysis of "United States Congress" where U.S. is a :mod of congress.
Should `:mode imperative` be used for command speech acts where there is no imperative sentence?
E.g., should I order you to yawn! be
(o / order-01
:ARG0 (i / i)
:ARG1 (y / you)
:ARG2 (y2 / yawn-01
:mode imperative
:ARG0 i))
or should :mode imperative
be omitted because the action ordered is not an imperative sentence?
It would be good to clarify this in http://www.isi.edu/~ulf/amr/lib/popup/mode.html
Document epistemic "must" in guidelines
See #28
Annotation of "therefore"
There have been different approaches to annotating "therefore" > infer-01 OR conclude-01 OR cause-01?
Depending on situation, can we go to the most appropriate choice?
China is a traditional ally and supplier of food and fuel to North Korea China and is therefore the country with the most leverage over North Korea.
(a / and
:op1 (s / supply-01
:ARG0 (c / country :name (n / name :op1 "China"))
:ARG1 (a3 / and
:op1 (f / food)
:op2 (f2 / fuel))
:ARG2 (c2 / country :name (n2 / name :op1 "North" :op2 "Korea")))
:op2 (a2 / ally-01
:ARG1 c
:ARG2 c2)
:ARG0-of (c3 / cause-01
:ARG1 (l / leverage-00
:ARG0 c
:ARG1 c2
:mod (m / most)))
:mod (t / tradition))
The issue therefore does not concern some
opinions that provoke waves of objection , but rather it fundamentally
concerns a vision of the role of the Church , its ideology , and its
universal message .
(i3 / infer-01
:ARG1 (c / contrast-01
:ARG1 (c2 / concern-02 :polarity -
:ARG0 (i / issue)
:ARG1 (t / thing
:ARG1-of (o / opine-01)
:ARG0-of (p / provoke-01
:ARG1 (w / wave
:consist-of (o2 / object-01)))
:mod (s / some)))
:ARG2 (c3 / concern-02
:ARG0 i
:ARG1 (v / view-02
:ARG1 (a / and
:op1 (r / role
:poss (r2 / religious-group :name (n / name :op1 "Church")))
:op2 (i2 / ideology
:poss r2)
:op3 (m / message
:mod (u / universal)
:poss r2)))
:manner (f / fundamental))))
Therefore , I call upon the citizens in general
, and those of al - Matn in particular , to show their support for the
Lebanese Army through civilized and democratic behavior as is their
habit in this area .
(c / conclude-01
:ARG1 (c2 / call-12
:ARG0 (i / i)
:ARG1 (a / and
:op1 (c3 / citizen
:mod (g / general))
:op2 (c4 / citizen
:location (c5 / city :name (n / name :op1 "at-Matn"))
:mod (p / particular)))
:ARG2 (s / show-01
:ARG0 a
:ARG1 (s2 / support-01
:ARG0 a
:ARG1 (a2 / army
:mod (c6 / country :name (n2 / name :op1 "Lebanon"))))
:manner (b / behave-01
:ARG0 a
:ARG1 (a3 / and
:op1 (c7 / civilization)
:op2 (d / democracy))
:mod (h / habit
:poss a
:location (a4 / area
:mod (t / this)))))))
Annotation of role/office/function as NE?
So far we have annotated semantically the job titles/functions occupied by a person, etc by decomposing it (e.g. (s/ secretary :mod (f/ foreign)). In this set we have come across some examples where it seemed somewhat odd to attempt to "decompose" a "job title":
- Secretary-General
- Secretary General
- Deputy Secretary General
- Secretary of State
- Foreign Secretary
- Foreign Press Secretary
- Chief Secretary
- State Public safety Secretary
- Secretary General of National Defense
- Home Secretary
- Deputy Minister of Foreign Affairs
- Prime Minister-in-waiting
To some extent, we thought that these might be annotated as NE of a certain position/function occupied by a person as some point in time.
Annotation of indefinite pronoun "one"
We had some sentences like :
- " Millions of those little objects , " he said , " which one sometimes sees in the sky . "
- " How is it possible for one to own the stars ? "
- "Whenever I met one of them who seemed to me at all clear - sighted , I tried the experiment of showing him my Drawing Number One , which I have always kept . "
In the majority of cases, "one" was annotated with the concept "one" unless we had to represent a subset relation, like in the #3, where "one" was annotated as "person".
Should we then not normalize the uses of "one" in sentences #1, #2 above to "person" as well for consistency reasons?
Epistemic "must"
How should we handle cases of "must" that aren't deontic (permission, obligation), but rather are epistemic? For example, "John isn't here, he must be sick."
We only have cases of "must" that correspond to "obligate," but that doesn't fit here. We considered using "probable" but "John isn't here, he MAY be sick" doesn't seem to quite capture the same thing.
This could also occur with other modals "He's gotta be/has to be sick if he's not here," etc.
:poss in consensus
-
consensus isi_0001.28 (snt. 27 in workset guidelines, last updated on Mon May 14, 2012)
naval officials(o / officer :poss (n / navy))
I would've used
:employed-by
or:mod
here, definitely not:poss
. -
consensus isi_0002.170 (snt. 207 in workset guidelines, last updated on Mon Mar 4, 2013)
academic year 2011-2012(d / date-entity :year 2011 :year2 2012 :calendar (y / year :poss (a / academia)))
Suggestion:
:mod
-
consensus isi_0002.171 (snt. 208 in workset guidelines, last updated on Mon Mar 4, 2013)
United States government fiscal year 2012(d / date-entity :year 2012 :calendar (y / year :poss (f / finance) :mod (g / government-organization :ARG0-of (g2 / govern-01 :ARG1 (c / country :name (n / name :op1 "United" :op2 "States"))))))
Suggestion:
:mod
-
Part/area of factory. Suggestion:
:part-of
- consensus nw.wsj_0003.25 (snt. 28 in workset wsj100-sent, last updated on Thu Apr 26, 2012)
Areas of the factory were particularly dusty where the crocidolite was used .
(d / dusty :domain (a / area :poss (f / factory) :location-of (u / use-01 :ARG1 (c / crocidolite))) :mod (p / particular))
- consensus nw.wsj_0003.27 (snt. 30 in workset wsj100-sent, last updated on Thu Aug 16, 2012)
Workers described `` clouds of blue dust '' that hung over parts of the factory , even though exhaust fans ventilated the area .
(d / describe-01 :ARG0 (p / person :ARG0-of (w / work-01)) :ARG1 (c3 / cloud :consist-of (d3 / dust :mod (b2 / blue)) :ARG1-of (h / hang-01 :location (o / over :op1 (p2 / part :poss (f / factory))) :concession (v / ventilate-01 :ARG0 (f2 / fan :mod (e / exhaust)) :ARG1 (a / area)))))
- consensus nw.wsj_0003.25 (snt. 28 in workset wsj100-sent, last updated on Thu Apr 26, 2012)
-
End of time period. Suggestion:
:part-of
-
consensus nw.wsj_0011.5 (snt. 92 in workset wsj100-sent, last updated on Fri Jul 13, 2012)
Government officials said exports at the end of the year would remain under a government target of $ 68 billion .
(s / say-01
:ARG0 (o / official
:mod (g2 / government-organization
:ARG0-of (g / govern-01)))
:ARG1 (r / remain-01
:ARG1 (t / thing
:ARG1-of (e3 / export-01)
:time (e2 / end
:poss (y / year)))
:ARG3 (u / under
:op1 (m2 / monetary-quantity :quant 68000000000
:unit (d2 / dollar)
:ARG1-of (t2 / target-01
:ARG0 g2)))))
- consensus wb.eng_0003.23 (snt. 23 in workset wb-eng-0003, last updated on Tue Mar 26, 2013)
Before the end of the decade , we may well see the unfunded liability zoom well past the original $ 400 million it took to build the original highway .
(p / possible
:domain (s / see-01
:ARG0 (w / we)
:ARG1 (z / zoom-01
:ARG1 (l / liability
:ARG1-of (f / fund-01 :polarity -))
:ARG4 (p2 / past
:op1 (m / monetary-quantity :quant 400000000
:unit (d / dollar)
:ARG1-of (t / take-10
:ARG0 (b2 / build-01
:ARG1 (h / highway
:mod (o / original))))
:mod (o2 / original))
:mod (w2 / well)))
:time (b / before
:op1 (e / end
:poss (d2 / decade))))
:degree (w3 / well))
-
Highway's exits. Suggestion:
:part-of
- consensus wb.eng_0003.5 (snt. 5 in workset wb-eng-0003, last updated on Mon Sep 10, 2012)
Already it is changing development patterns for the worse and causing localized traffic congestion around its exits .
(a / and :op1 (c / change-01 :ARG0 (i / it) :ARG1 (p / pattern :mod (d / develop-02)) :ARG2 (b / bad :degree (m / more))) :op2 (c2 / cause-01 :ARG0 i :ARG1 (c3 / congest-01 :ARG1 (a2 / around :op1 (e / exit :poss i)) :ARG2 (t / traffic) :ARG1-of (l2 / localize-01))) :time (a3 / already))
- consensus wb.eng_0003.20 (snt. 20 in workset wb-eng-0003, last updated on Mon Mar 25, 2013)
If Route 288 has already created a $ 13 million road - construction in just this one exist , think of the liability it has created for all of its exits !
(t / think-01 :mode imperative :ARG0 (y / you) :ARG1 (l / liable-41 :ARG1-of (c / create-01 :ARG0 r2 :ARG3 (e / exit :mod (a / all) :poss r2))) :ARG1-of (c5 / cause-01 :ARG0 (c2 / create-01 :ARG0 (r2 / road :name (r / name :op1 "Route" :op2 "288")) :ARG1 (c3 / construct-01 :ARG1 (r3 / road) :ARG1-of (c4 / cost-01 :ARG2 (m / monetary-quantity :quant 13000000 :unit (d / dollar))) :location (e2 / exit :quant 1 :mod (t2 / this) :mod (j / just))) :time (a2 / already))))
- consensus wb.eng_0003.5 (snt. 5 in workset wb-eng-0003, last updated on Mon Sep 10, 2012)
-
Shares of stock. Suggestion:
:consist-of
- consensus wsj_1129.1 (snt. 26 in workset first50, last updated on Fri Apr 19, 2013)
Brush Wellman Inc. said its board increased the number of shares of common stock to be purchased under a previously authorized program to 3.9 million from 2.9 million .
(s / say-01 :ARG0 (c2 / company :name (n5 / name :op1 "Brush" :op2 "Wellman" :op3 "Inc.")) :ARG1 (i / increase-01 :ARG0 (b / board :poss c2) :ARG1 (n / number :quant-of (s2 / share :poss (s3 / stock :ARG1-of (p2 / purchase-01 :prep-under (p3 / program :ARG1-of (a / authorize-01 :time (p4 / previous)))) :mod (c / common)))) :ARG3 2900000 :ARG4 3900000))
- consensus wsj_1129.1 (snt. 26 in workset first50, last updated on Fri Apr 19, 2013)
Date-entities - dayperiod
Should all these expressions like: "sunrise, dawn, morning, afternoon, noon, evening, dusk, sunset, night, midnight" be annotated as date-entities using a :dayperiod role?
There seem to be some contradicting annotations in consensus:
[2] consensus chtb_0271.3 (snt. 14 in workset first50, last updated on Sun Apr 22, 2012)
Signed here this morning were the two projects , knowledge information network communication technology and DNA bio-technology .
(s / sign-01
:ARG1 (p / project :quant 2
:mod (a / and
:op1 (t3 / technology
:instrument-of (c / communicate-01
:mod (n / network
:mod (i / information)
:mod (k / knowledge))))
:op2 (b / bio-technology
:mod (d / dna))))
:location (h / here)
:time (m / morning
:mod (t / this)))
[3] consensus nw.wsj_0010.9 (snt. 76 in workset wsj100-sent, last updated on Tue Aug 21, 2012)
The next morning , with a police escort , busloads of executives and their wives raced to the Indianapolis Motor Speedway , unimpeded by traffic or red lights .
(r / race-01
:ARG0 (a / and
:op1 (e / executive)
:op2 (w / wife
:poss e)
:quant (b / busload)
:ARG1-of (e2 / escort-01
:ARG0 (p / police)
:ARG4 s))
:ARG1 (s / sports-facility :name (i / name :op1 "Indianapolis" :op2 "Motor" :op3 "Speedway"))
:ARG1-of (i2 / impede-01 :polarity -
:ARG0 (o / or
:op1 (t / traffic)
:op2 (l / light
:mod (r2 / red))))
:time (d / date-entity
:dayperiod (m2 / morning)
:mod (n2 / next)))
Thank you.
annotation of membership
There was already an example for this is the guidelines (consensus isi_0002.16 (snt. 54 in workset guidelines, last updated on Tue Jul 3, 2012):
"26 of the 44 countries in Europe are members of NATO."
which is annotated as:
Wouldn't it be a better option to extend the scope of include-01 to cover these cases, rather than use :poss? As in:
"the new EU members"
OR
"Albanian officials aspire to EU membership."
NATIONALITY named entities
In my work to add general named-entity pre-processing to the AMR Editor, I noticed that "nationality"-type named entities are being annotated somewhat inconsistently with respect to the role connecting the GPE to "person":
Example: "This indicates the number of North Koreans working in Iran on such projects."
- (p / person :mod <North Korea>)
- (p / person :source <North Korea>)
- (p / person :poss <North Korea>)
- (p / person :location <North Korea>)
:mod appears to be the mildly dominant annotator choice:
(p / person
:mod (c / country :name (n / name :op1 "North" :op2 "Korea")))
and it probably has the most relevant precedents, so I'm leaning towards :mod for "nationality":
British conglomerate
(c / conglomerate
:mod (c2 / country :name (n / name :op1 "Britain")))
Canadian dollar
(d / dollar
:mod (c / country :name (n / name :op1 "Canada"))
Such nationality NEs are not limited to nationality in the legal sense, but would probably extend to non-country GPEs as well:
- Californians
- Londoners
- Asians
Deleting one ancestor of a reentrant variable causes loss of dependent information
(s / say-01
:ARG0 (p / prince
:mod (l / little)
:ARG0-of (t / turn-01
:ARG1 p
:purpose (g2 / give-01
:ARG0 p
:ARG2 (s2 / side
:part-of (p2 / planet)
:mod (o / other)))
:purpose (g / glance-01
:ARG0 p
:ARG1 s2
:quant (x / 1
:mod (m / more)))))
:ARG1 (c / contrast-01
:ARG2 (l2 / look-01
:ARG0 p
:time (a2 / already))))
== del g2
==>
(s / say-01
:ARG0 (p / prince
:mod (l / little)
:ARG0-of (t / turn-01
:ARG1 p
:purpose (g / glance-01
:ARG0 p
:ARG1 (s2 / side)
:quant (x / 1
:mod (m / more)))))
:ARG1 (c / contrast-01
:ARG2 (l2 / look-01
:ARG0 p
:time (a2 / already))))
The concept side
is correctly moved to the s2
under g
, but its modifiers (:part-of p2
and :mod o
) disappear.
Hyphenated NE
As a rule, in the case of hyphenated NE, should the 2 tokens "x-y" be annotated on the same op irrespective of the type of tokenization ?
Or should we consider the hyphen just another token and place it on a separate :op?
In consensus - Different approaches possibly due to different types of tokenization:
Examples:
Wi Sung-lac stated to reporters that Charles Kartman stated that the North Korean nuclear reactor project is linked to North Korea's problem with nuclear weapons.
p / person
:name (n / name
:op1 "Wi"
:op2 "Sung-lac"))
--> 2 or 3 :opx
The NCRI is the People's Mujahedeen's political front in Auvers-sur-Oise.
c / city
:name (n / name
:op1 "Auvers-sur-Oise"))
--> 1 or 5 opx?
High frequency phrase: cross-border
Should we dissect this into thing that crosses border, or leave as :mod cross-border? CJ consensus is using :mod, but it doesn't seem like too much of a stretch to provide more information by using the "cross" sense, meaning "move perpendicularly across a barrier. Here's an example with "cross-border piracy":
ADV -> ADJ -> N -> V
AMR stems down to representative concepts -- so, instead of "behavior", "behave", "behavioral", and "behaviorally", we just use "behave-01" in AMR. Then of course, there are many ways to realize (in English) an AMR that contains "behave-01".
The general plan is "ADV -> ADJ -> N -> V".
Between ADJ and N, some possible exceptions crop up. For example, I have seen AMR annotators render the word "patient" (English) as "patience" (in AMR) -- do we want to do this? I believe Ulf has seen several cases where the N is a morphological variant of the ADJ.
As long as we use one canonical concept, it doesn't matter, and we don't want to get caught up in English morphology. But we need a rule for deciding what the canonical concept is. "ADV -> ADJ -> N -> V" is one such rule, and we might decide to stick with it.
Rankings and comparisons
At CU we've had several instances come up that mention ranks of entities in comparison to others:
And also:
Sentence: Rashwan and other Egyptian experts said that Adel has become Al-Qaeda's number 3 in command since becoming chief of military operations and succeeding Abu Hafs al-Masri.
In the past, for temporal "first" we've just used :mod. We've considered a variety of options incorporating :li and comparisons, but we're just not sure how to handle these. It might be nice to introduce a :ranking concept.
Phrasal verbs: drop directional particles?
We have decided to use a simple, lexical verb entry where this only differs from a phrasal verb in aspectual qualities (e.g. eat vs. eat up). In some cases, e.g. lay down, come back, we can use the simple lexical verb, but should the particle be included at all? e.g. ":direction back/down"? Or should this be simplified into using just the lexical verb? I was including the directional particle in cases where roleset included :argm DIR or a numbered arg for direction (e.g. "come"), and leaving it out where the roleset didn't include this, (e.g. "sit/lay down").
"To me..." introductions
I have a couple of cases where "to you, ...." and "to me, ..." are used (e.g. "To you , I shall be unique in all the world ... "). I have left these as :prep-to you/me for lack of a better option. Did anyone else come up with something better for these?
Parentheticals with relevant content
I've come across a few examples of parentheticals within a sentence in the Proxy data that seem to carry more significant semantic content than what we have seen before. Here are some examples:
The attacks range from viruses, flame-mail (abusive e-mail), stone throwing (disfiguring a home page) to cyber squatters and scripts for breaking into Internet sites.
Threats and risks are managed like drinking water from (Indian) taps, and visiting Srinagar (in restive Kashmir).
What is the best way to treat this kind of information in an AMR? In the first example they are giving examples/defining the term they follow; in the second example they seem to be giving information that might be unsure or extraneous.
"All" :mod or :quant?
I have noticed some inconsistency in how "all" is used, sometimes as :mod and sometimes as :quant. I think it can genuinely be both, perhaps :mod where it means "completely" (e.g. It's all done) and :quant where it indicates the entire quantity of something (e.g. "all the boys went"). But I wanted to make sure this seemed correct, and I think there can be some difficult borderline cases (e.g. all the chickens are just alike, vs. the chickens are all alike). I've used :quant all for the following phrase:
All the chickens are just alike , and all the men are just alike .
verbalize "effort"?
I believe that we would map the light verb construction "make effort" to "try," but should we map all mentions of "effort" to "try"? For example, in mentions like "as part of an effort," or "during efforts..." These mentions are often accompanied by who is making the effort and what is being tried, so the argument structure would be very useful.
"Effort" has come up quite a bit in proxy, it seems that it is generally being treated as a noun, with the thing tried marked as :purpose.
VERB -> ADJ
One of our new summer interns, Afton Coombs, suggested to map causative verbs to "make" + adjective, e.g.
- sadden -> make sad
- simplify -> make simple
This would help to map more sentences with the same meaning to the same AMR, and make the core of word pairs such as sad/sadden semantically equivalent.
We already do not to verbalize the adjectives that causative verbs are based on:
- DO-NOT-VERBALIZE familiar TO familiarize-01
- DO-NOT-VERBALIZE liquid TO liquefy-01
- DO-NOT-VERBALIZE sad TO sadden-01
- DO-NOT-VERBALIZE simple TO simplify-01
- DO-NOT-VERBALIZE stable TO stabilize-01
- DO-NOT-VERBALIZE white TO whiten-01
VERB -> ADJ mappings would be an extension of our recently discussed NOUN -> ADJ mappings (e.g. sadness -> sad). In both cases the reverse POS mapping directions ADJ -> VERB and ADJ -> NOUN still dominate (e.g. reliance/reliant -> rely-01).
Current annotation (per guidelines) for "The girl was saddened by the disaster."
(s / sadden-01
:ARG1 (g / girl)
:ARG2 (d / disaster))
Proposed annotation:
(m / make-02
:ARG0 (d / disaster)
:ARG1 (s / sad
:domain (g / girl)))
Scope: At least for the time being, this proposal is limited to VERB -> make-02 + ADJ. It would not include more complex mappings such as hospitalize -> admit/put/send to a hospital.
I looked into how often this might occur in the corpus. Actually, a bit more often than I first thought, if we include verbs such as open, strengthen, modernize, publicize, clarify.
We would add the more common such mappings to our AMR lists so that the AMR Editor could provide guidance to annotators on specific words.
Choices include: (1) no change, (2) add to guidelines now, (3) hold off until AMR 1.5
Use of quantities as generic entity types?
This question came up in training: can we use things like "monetary-quantity" for example, as a generic entity type for cases where the monetary quantity is implicit?
It is used that way a couple of times in consensus. In the first example below, one monetary quantity is explicitly mentioned, but the other is implicit. In the second example, the only mention is "spending," but monetary-quantity is used to represent the implicit concept of money.
Can we generally assume that these quantities, and perhaps even other entities like date-entity, are also usable as generic entity types? Or should this practice be avoided? I know we've had discussion about not using quantity types when it's not a specific, measurable quantity, so I wasn't sure if these were oversights, or something we could put into practice.
Coreference difficulties
I have several phrases like "We are roses" said the roses; "I am a fox" said the fox.
I'm not sure how to treat the coreference in these sentences, or whether we want some kind of subset relation? This seems like the kind of case where we WOULDN'T want to use a subset relation since the superset isn't actually mentioned. If they are all coreferring (we, roses, roses; I, fox, fox), the AMR representation seems very strange to me. Here's what I have now:
(s / say-01
:ARG0 (r / rose)
:ARG1 (r2 / rose
:domain r))
simply leaving two different rose entities that aren't coreferring...
And, just to double check, in cases of quotes like this, we wouldn't introduce a pronoun "we", correct? We maintain the coreference with the more specific referent, like "rose" across the quote boundary? We might want to put an example like this in the guidelines, because many of my annotations didn't make the entity mention inside and outside the quote coreferential.
"or" as an alternative way of expressing paranthetical equivalence
If there is a case of "translation" from one scale into another, with the use of "or" that works like a "parenthetical", should the 2 alternative be dropped?
e.g.
about 2800 kilometers or 1740 miles east of Moscow
The deal states Iran will supply 5.5 billion cubic meters or 194 trillion cubic feet of gas annually from 2011.
If we had "about 2800 kilometers (1740 miles) east of Moscow", we would not annotate "1740 miles". Should the same principle be applied here?
New NE types: ideology and sport?
- NE: "ideology"
In the example "Zionist and American plans to eliminate the Islamic Movement", we did not find a good fit in terms of NE type and currently annotated it as:
(t / thing :name (n2 / name :op1 "Zionism"))
However, should we consider adding a new NE type: movement/ideology, and in this case should it cover concepts like: communism, democracy, socialist, etc.
Or should we annotate them as a common noun?
- "sport"
We came across examples like:
"tai chi , and meihuazhuang, drunken fist" which are different types of martial arts and we did not find a good NE type.
The "sport" NE type could cover most types of sports activities.
Replace concept command should be consistent with reentrant variables
Right now rc
can modify the variable as well as the concept. However, it only replaces the occurrence of the variable that is next to the concept. For example:
(d / difficult
:degree (m / more
:degree (m2 / much))
:domain (j / judge-01
:ARG0 (y / you)
:ARG1 y)
:compared-to (j2 / judge-01
:ARG1 (o / other)))
== rc y one
==>
(d / difficult
:degree (m / more
:degree (m2 / much))
:domain (j / judge-01
:ARG0 (o2 / one)
:ARG1 y)
:compared-to (j2 / judge-01
:ARG1 (o / other)))
This results in a variable (y
) which now has no concept.
some questions on the amr tutorial sentences
1.-----
The boy must not go. [isi_0001_5]
Annotator archna
(o / obligate-01
:ARG1 (b / boy)
:ARG2 (g / go-02
:polarity -
:ARG0 b))
Annotator consensus
(o / obligate-01
:ARG2 (g / go-02
:polarity -
:ARG0 (b / boy)))
QUESTION: Why do we not have an ARG1 for "obligate" - is it because we assume that we don't have 2 clauses here?
2.-----
The man died between the house and the river. [isi_0001_21]
Annotator archna
(d / die-01
:ARG1 (m / man)
:location (b / between
:op1 (h / house)
:op2 (r / river)))
Annotator consensus
(d / die-01
:ARG1 (m / man)
:location (b / between
:op1 (h / house)
:op1 (r / river)))
QUESTION: Is there a principled way to determine if we will get the same relation op1 or different op1, op2 etc? In the following we get two different role names.
[35] consensus isi_0002.95 (snt. 132 in workset guidelines, last updated on Tue Jul 3, 2012)
The performance will take between 45 minutes and an hour.
(p / perform-01
:duration (b / between
:op1 (t / temporal-quantity :quant 45
:unit (m / minute))
:op2 (t2 / temporal-quantity :quant 1
:unit (h / hour))))
3.-----
He was a participant in the scheme. [isi_0001_22]
Annotator archna
(p / participate-01
:ARG0 (h / he
:ARG2-of (s / scheme-01)))
Annotator consensus
(p / participate-01
:ARG0 (h / he)
:ARG1 (s / scheme))
QUESTION: Why is "scheme" ARG1 of "participate"?
4.-----
According to government sources, the killing happened yesterday. [isi_0001_23]
Annotator archna TOP: report-01
(r / report-01
:ARG0 (s / source
:mod (g / government-organization))
:ARG1 (k / kill-01
:time (y / yesterday)))
Annotator consensus TOP: say-01
(s / say-01
:ARG0 (s2 / source
:mod (g / government-organization
:ARG0-of (g2 / govern-01)))
:ARG1 (k / kill-01
:time (y / yesterday)))
QUESTION: Why do we have ARG0-of? Is that necessary?
5.-----
The boy came and left. [isi_0001_24]
Annotator archna
(a / and
:op1 (c / come-01
:ARG0 (b / boy))
:op2 (l / leave-01
:ARG0 b))
Annotator consensus
(a / and
:op1 (c / come-01
:ARG1 (b / boy))
:op2 (l / leave-01
:ARG0 b))
QUESTION: Why is "boy" Arg1 instead of Arg0 for the predicate "come"? And do we differentiate between the unaccusative and unergative intransitive verbs for the PB labels here?
6.-----
The boy is a hard worker. [isi_0001_25]
Annotator archna TOP: work-01
(w / work-01
:ARG0 (b / boy)
:manner (h / hard))
Annotator consensus TOP: person
(p / person
:domain (b / boy)
:ARG0-of (w / work-01
:manner (h / hard)))
QUESTION/COMMENT: I thought we would use "top person" when we want to express a noun phrase like "a hard worker (boy)" instead of a clause. If we have two ways to express this semantic content, we need to be consistent in whether we use one form or the other, I think we choose that depending on what is being focussed. If it is a nominal expression "a hard worker (boy)", the person is being focussed and the second annotation above seems better; but if it is clausal, the fact that the boy is a hard worker is focussed, so we use the first annotation above.
7.-----
the earlier plan
(p / plan
:time (e / early
:degree (m / more)))
QUESTION: This question is not that important but why do we not mark it as "plan-01"? Is it because it was not a verb but a noun?
8.-----
several hundred boys [isi_0001_32]
Annotator archna
(b / boy
:quant (h / hundred
:mod (s / several)))
Annotator consensus
(b / boy
:quant (s / several
:op1 100))
QUESTION: I'm not sure about this difference. I looked at the guidelines, I see we do it this (consensus) way, but I am just curious why we do it this way.
9.-----
about 10 miles [isi_0001_35]
Annotator archna TOP: distance-quantity
(d / distance-quantity
:quant (a / about
:op1 10)
:unit (m / mile))
Annotator consensus TOP: about
(a / about
:op1 (d / distance-quantity
:quant 10
:unit (m / mile)))
QUESTION: What is the difference between "about 10 miles" and "more than 4000 boys"/ "between 4000 and 5000 boys"? All are variations of quantities.
:degree inconsistency
[sentence numbers are w.r.t. consensus search for :degree]
- comparative/superlative adj/adv: 1, 2, 3, 4, 6, 8, 11, 12, 13, 15, 16, 17, 18, 19, 21, 22, 23 sooner, ...
- degree modifier: intensifier/downtowner other than ‘more’ or ‘most’ before adj stem: 5 too, 7 very, 10 extreme(ly), 14 very, 29 less(er), 36 very, 37 much, 39 dead
- hedge: 35 large(ly)
- epistemic hedge?: 33 we may well see
- anti-hedge?: 43 anything remote(ly) resembling
- extent: 9 further, 20 further
- quantity/intensity: 40 the least stress, 41 maximum use, 42 more places, 44 heavy demands
- isi_0002.144 more police officers: :quant more
- nw.wsj_0003.19 most schools: :quant most
- wb.eng_0003.10 more money: :mod more
- wb.eng_0003.49 more money: :quant more
- wb.eng_0003.61 more of our transportation bill: [bill-01 :quant more]
- wb.eng_0003.61 more travel: [travel-01 :mod more]
- wb.eng_0003.77 more places: :quant more
- (I think :quant seems most straightforward for these cases.)
- treat more like X than Y: 28
45 “police are facing more rowdy behavior”—ambiguous! the AMR has
:manner [rowdy :degree more]
suggesting “rowdier”. But “greater amount of rowdy behavior” should be :quant?
NB: “more than” is consistently more-than
Approximately
We noticed some inconsistencies related to the annotation of approximately.
The 2 approaches have been:
- to go to "approximate-01" and make use of the roles
"cost approximately 51 billion rupees"
:ARG2 (m2 / monetary-quantity
:unit (r / rupee)
:quant 51000000000
:ARG1-of (a3 / approximate-01))))
- :quant (a / approximately :op1 )
Reification of :quant
We need a reification of :quant, for "The number of enterprises totaled 5883".
need-01 versus necessary versus obligate-01
We use obligate-01 for "must" and "obliged to". Do we need "need-01" sometimes?
They always need to have things explained .
(n / need-01
:ARG0 (t / they)
:ARG1 (e / explain-01)
:time (a / always))
Review of AMR Checker lists
In order to support annotators in improving annotation consistency, the AMR Checker uses a number of lists, including some that I have built or augmented for AMR. Here's the primary such resource of lists:
http://www.isi.edu/~ulf/amr/lib/amr-auto-correction-rules.txt
You are invited to have a look at it and provide feedback.
Many of the entries were extracted from our growing corpus of AMRs. For my decisions (e.g. VERBALIZE vs. DO-NOT-VERBALIZE) I took into consideration of what LDC and other annotators have come up with. The list of 150 mappings from support verb constructions to AMR concepts also greatly benefitted a list that Claire kindly sent me.
Below are a few particular cases where I was not sure, as indicated by a comment-out (#) or a MAYBE- prefix. For each class of advice, I also include a few normal cases in order
for you to get a quick sense of the class.
DO-NOT-VERBALIZE professor TO person :ARG0-of profess-01
DO-NOT-VERBALIZE white TO whiten-01
DO-NOT-VERBALIZE relationship TO relate-01
DO-NOT-VERBALIZE suspicious TO suspect-01
DO-NOT-VERBALIZE valuable TO value-01
DO-NOT-MAP international TO internation
DO-NOT-MAP friendly TO friend
DO-NOT-MAP official TO officer
DO-NOT-MAP various TO variety
VERBALIZE afraid TO fear-01
VERBALIZE approval TO approve-01
VERBALIZE favorable TO favor-01
VERBALIZE hungry TO hunger-01
VERBALIZE scandalous TO scandalize-01
VERBALIZE speedy TO speed-01
VERBALIZE dinner TO dine-01
MAYBE-VERBALIZE terrorism TO terrorize-01
Why did we decide against a pianist being a piano player?
VERBALIZE drummer TO person :ARG0-of play-01 :ARG1 drum
VERBALIZE pianist TO person :ARG0-of play-01 :ARG1 piano
VERBALIZE trumpeter TO person :ARG0-of play-01 :ARG1 trumpet
DO-NOT-VERBALIZE chairman TO person :ARG0-of chair-01
DO-NOT-VERBALIZE editor TO person :ARG0-of edit-01
DO-NOT-VERBALIZE president TO person :ARG0-of preside-01
Not 100% sure about the following:
MAYBE-VERBALIZE commander TO person :ARG0-of command-02
Corpus usage: military commanders
MAYBE-VERBALIZE captor TO person :ARG0-of capture-01
Corpus usage: hostage taking, kidnapping
MAYBE-VERBALIZE container TO product :ARG0-of contain-01
Corpus usage: shipping containers, container cranes
MAYBE-VERBALIZE carrier TO ship :ARG0-of carry-01
Corpus usage: aircraft carrier
MAYBE-VERBALIZE contractor TO person :ARG0-of contract-02
Corpus usage: defense contractor, insurance contractor
MAYBE-VERBALIZE devotee TO person :ARG0-of devote-01
Corpus usage: Buddhism, charities
MAYBE-VERBALIZE terrorist TO person :ARG0-of terrorize-01
MAP Kirghizia TO Kyrgyzstan
MAP natural TO nature
MAP worthy TO worth
SPLIT moonlight TO light :source moon
SPLIT moonlight TO light-02 :ARG0 moon
Mass destruction
In full searches, I've noticed some inconsistencies in the common phrase (for proxy) "weapons of mass destruction." The good news is that these are regularly getting verbalized as destroy-01, but what is "mass"? Some use Arg1, some Manner, and quite a few use Mod:
:ARG1 (w / weapon
:ARG2-of (d / destroy-01
:ARG1 (m4 / mass)))))
:ARG1 (w2 / weapon
:ARG2-of (d2 / destroy-01
:manner (m2 / mass)))))))))
:ARG1 (w2 / weapon
:instrument-of (d2 / destroy-01
:mod (m / mass))))))))))
I actually like :degree for mass. This phrase is common enough that I think we should set up a consensus representation for it. Apologies, I can't remember how to format these with the correct indentation in Github.
Questions about "it" and resemble-01
I've seen a few sentences that are structured something like "It will be as it was with X" and am just not sure how to best treat them in an AMR. Here are the examples I've seen so far:
"It is just as it is with the water."
(r / resemble-01
:ARG1 (i / it)
:ARG2 (w / water)
:mod (j / just))
It will be as it was when we drank the water ... "
(r / resemble-01
:ARG1 (i / it)
:ARG2 (t / time
:time-of (d / drink-01
:ARG0 (w / we)
:ARG1 (w2 / water))))
I don't speak French well, but these "it is just as it is with the X" sentences sound like a poetic translation. In fact, the French line is "C'est comme pour la fleur." Which I think is more literally like "It is like the flower." I don't know what to call that "it". It's like saying "Take the flower, for example" in English.
Annotation of ambigous agent types
If it's not clear whether the agent of action is a person or company etc., should we use thing or rely on the context?
e.g.: China was to have announced a winner by the end of 2005 but has delayed its decision until early next year hoping to elicit better terms from the bidders.
A proposal for negatives with quantifiers
- No one is sleeping. [negative has scope over "one".]
(s / sleep
:ARG0 (o / one :polarity -))
- Someone is not sleeping. [negative as scope over "sleep" if "there exists someone who is not sleeping" with focus/emphasis on "not".]
(s / sleep :polarity -
:ARG0 (s2 / someone))
- Someone is not sleeping. [negative as scope over "someone" if "there exists someone who is not sleeping" with focus/emphasis on "someone".]
(s / sleep
:ARG0 (s2 / someone :polarity -))
- Everyone is not sleeping. [meaning: "some are sleeping, some are not.", i.e. emphasis on "not"; "not" has scope over the verb "sleep"]
(s / sleep :polarity -
:ARG0 (a / all))
- Everyone is not sleeping. [meaning: "no one is sleeping.", i.e. emphasis on "everyone", "not" has scope over the quantifier "everyone"]
(s / sleep
:ARG0 (a / all :polarity -))
- He sometimes does not sleep. [negative has scope over the verb "sleep"]
(s / sleep :polarity -
:ARG0 (h / he)
:frequency (s2 / sometimes))
- He never sleeps. [negative has scope over the verb "sleep"]
(s / sleep :polarity -
:ARG0 (h / he)
:time (e / ever))
- I like nobody. [negative scopes over "one".]
(l / like
:ARG0 "I"
:ARG1 (o / one :polarity -))
- I don't like anybody. [negative scopes over "one"/"anybody", that is why same AMR as above.]
(l / like
:ARG0 "I"
:ARG1 (o / one :polarity -))
- I don't like somebody. [Like in the 5 and 6 cases above, negative scopes over the verb "like".]
(l / like :polarity -
:ARG0 "I"
:ARG1 (o / one))
- I don't like everybody. [meaning: "some I like, some I don't."; negatives scopes over the verb "like"]
(l / like :polarity -
:ARG0 "I"
:ARG1 (a / all))
- I don't like everybody. [when this meaning is possible: "no one who I like"; negative scopes over the quantifier "everyone"]
(l / like
:ARG0 "I"
:ARG1 (a / all :polarity -))
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.