|
Multimodal Information
Group Home
Benchmark Tests
Tools
Test
Beds
Publications
Links
Contacts
|
ACE Phase 1 (2000)
EDT Guideline Clarifications
Prepared March 9-10, 2000
with corrections to 2c and 3 made March 12, 2000
based on a presentation to the EDT community, March 9, 2000
A. Meyers and R. Grishman, NYU
1. Extents
It is important to include all modifiers in the extent, even if they are
non-restrictive modifiers or modifiers of the modifiers.
1a. Floated quantifiers
A floated quantifier is a quantifier, such as "both", "all", or "each" which
modifies a noun phrase but appears in a position other than at the beginning
of the noun phrase; for example, "Fred and Mary both like parsing."
Floated quantifiers should not be included in the extent of the noun phrase
they modify. Thus in
- [He] and [the pope] both want to end the exploitation of man by man.
the extents are "he" and "the pope".
1b. Identification of announcers in broadcast non-sentences
In the phrase
- "[Pamela McCall] in [the [BBC] news room in [London]]"
the extent of the announcer mention is just "Pamela McCall"; the
following phrases are not part of the mention of the announcer.
This would also be the case if the text was
- "[Pamela McCall], BBC News, [London]"
Similarly for "Orelon Sidney with your latest headline news weather update",
the mention of the announcer is just "Orelon Sidney".
2. GSPs
2a. "Arab"
- "... she will discuss this situation with [Arab] leaders." (CNN 599)
"Arab" is not an adjective associated with a GSP [it is not associated
with a specific political unit or set of units]
2b. "the French"
- "I'm still working with uh Russians, [the French], the United Nations"
Treating words like "the French", "the English", "the Dutch" when they
refer to a group of people: they are instances of plural nouns
which evoke entities of type person. Thus, in this example, "the
French" evokes an entity of type person with extent "the French" and head
"French". This contrasts with anaphoric usage, such as "You
like the German wine, but I like the French." In such a case, "French"
is a proper adjective which evokes a GSP entity.
2c. plural GSPs
A mention of a set of GSPs, in the form of an NP whose head is a plural
GSP noun, evokes an entity of type GSP. Thus the phrase "the southern
states" evokes an entity of type GSP. In contrast, the phrase "the
South" evokes an entity of type location. The difference between
these cases is the type of the head of the phrase.
3. Named heads of unmarkable phrases
If a full noun phrase is not markable by EDT rules, and the head of the
phrase is a name which, if it stood alone, would evoke an EDT entity, then
treat the name as a mention whose extent and head are this name.
- "Today's [New York Times]" (NYT 485)
In this example, "Today's New York Times" refers to an issue of a newspaper,
which is not a reportable entity. However, the head of the phrase,
"New York Times", is a name which is a mention of an entity of type organization.
Therefore we generate a mention of this entity, whose head and extent
are "New York Times".
- "The next [Jordan]? Not Yet..." (NYT 471)
Here "Jordan" does refer to a real person, but the full noun phrase means
something like "the next person like Jordan" and in the current context
is an attribute which is not being positively asserted (of Kobe Bryant).
It is therefore not markable (generic). However, the head, "Jordan",
is markable as a mention of the entity Michael Jordan.
4. Periodicals are organizations
Following the named entity guidelines (NE99, section 5.4.4), all periodicals
are to be treated as organizations. We don't require annotators to
know which periodicals correspond to organizations of the same name
- The Archives of Disease in Childhood (CNN 1006)
- The New York Times
Both of these are to be treated as organizations.
5. Names
As the guidelines indicate, names are atomic ... names within other names
are not to be marked. It is therefore important to know what constitutes
a name, even if the name is not annotated (because it is not one of the
EDT types).
Law cases are not names, so the individual constituents (the plaintiff
and defendant) are markable:
- [Barry] vs. [the United States] (NYT 485)
Holiday names are names, so "Valentine" is not markable (as a person
entity) in
- Valentine's Day (Minicorpus 1 APW 467)
Speech names, such as "State of the Union" (Minicorpus 1:ea 95), are names,
so the constituent "the Union" is not markable.
Titles are not names, so in "Secretary of [state]" (CNN 599) "state"
can be annotated as a reference to an organization (the Department of
State).
6. Facility vs. organization
Many entities, such as museums and schools, have characteristics of both
facilities and organizations. In some cases, we explicitly list in
the guidelines whether such entities should be typed as facilities or as
organizations.
For other ambiguous entities which are not explicity listed, we provide
the following general guidance: if the class of facility/organization
evokes in the listener an image of a building (e.g., a hospital or a hotel),
then classify the entity as a facility.
7. TV Programs
A TV network or station is an organization, but a TV program is not. If
a TV network name is embedded in a program name (ABC News, CBS Nightly News),
it is not markable since names are indivisible.
Although some program names may also be names of organizations, this
may be difficult for the annotator to ascertain, so we uniformly consider
TV program names not markable.
8. Organization/person
If a text does not make clear if an entity is an organization or a set of
people (without formal association), mark the entity as a person.
In all three cases below, the bracketed item is to be marked as a person
entity.
- "[Animal advocates] _ with a big assist from the pet food lobby _
say these problems should be dealt with on a case-by-case basis."
(NYT 486)
- "[Authorities] have just used water on the fire for the first time
and they say there was a risk the impact of the water on four hundred
tons of burning ammonium nitrate, might have caused an explosion." (mini-1
CNN 3)
- "[Authorities] said the new casualties included wounded people who
later died in hospital and bodies of victims taken from the scene before
being accounted for." (mini-1 APW 484)
9. "We/Us/Our" in news broadcast articles
"We/Us/Our" in news broadcast articles are to be marked as mentions of entities
of type person. They are considered to refer to the staff of the News
show, and not to be coreferential with the broadcasting organization (ABC,
CBS).
"Just before [we] leave you, a brief review of the
day's two major stories." (ABC 1676)
"A couple of minutes ago, [we] reported that the big
tobacco companies were in Congress today, pushing hard for a national
settlement of al l the lawsuits against them. (ea 809)
"Finally from [us] this evening, [our] regular Monday
report" (ea 1564)
10. "The World"
"The world" is frequently used to represent the entire population of the
planet or all the governments, etc. It is markable as a mention
of an entity of type location.
"[The entire world] will see images of the Pope in
Cuba." (ea 877)
11. Generic vs. specific
11a. Indefinite NPs that are reported as part of concrete events
should be interpreted as specific (nongeneric), even in cases of plural
NPs of nonspecific number. Consider
"[Animal advocates] _ with a big assist from the pet food
lobby _
say these problems should be dealt with on a case-by-case
basis."
(NYT 486)
If you believe that an actual set of animal advocates are taking this
position, then it is nongeneric. If you believe that the reporters are
making a generalization about animal advocates around the country or throughout
the entire world, then it should be interpreted as generic.
If the text read "said" (instead of "say"), it would probably be referring
to a specific event and so would be a nongeneric NP.
11b. If there is a literal interpretation of a sentence with an
NP referring to a specific set, the NP is markable, even if there is an
implied reading in which the NP is generic (does not refer to a specific
set). For example, in
- "All the musicians that I've known are drunks."
The speaker is making a statement about a finite set (all the musicians
that I've ever known). The audience may be expected to generalize this
set to all musicians generically, but the literal reading is specific,
so this mention is reportable.
11c. If there is a literal interpretation of a sentence
with an NP referring to a generic set, the NP is generic (non-markable),
event if there is an implied coreference of this (overtly generic) NP
with another entity mentioned in the discourse. Thus in
- Fiercely agressive representation ... cannot be an excuse for smearing
[a lawyer] (NYT 463)
"a lawyer" is to be treated as generic (non-markable), even though there
is an implied reference to a specific lawyer mentioned earlier in this
discourse.
[ ACE Home ]
Page Created: September 6, 2007
Last Updated: November 4, 2008
|
ACE Phase 1 links:
ACE Phase 1 Home
Documentation
Schedule
Resources
Contacts
ACE Home
|
