TimeML Annotation Guidelines

Annotation Guideline Version: 0.1

Release Date: June 4, 2002

Authors: Roser Saurí, James Pustejovsky, Bob Ingria

TERQAS Annotation Working Group Members: Lisa Ferro, Marcia Lazo, David Day, Patrick Hanks, Marc Verhagen, Roser Saurí, José Castaño, Bob Ingria, James Pustejovsky.

Contents

  1. Introduction
  2. Explanation of TimeML Tags
    2.1 The tag <EVENT>
           2.1.1 How to annotate EVENTs
           2.1.2 Attributes for EVENTs
           2.1.3 BNF for the EVENT tag
           2.1.4 Complete annotation of some of the previous examples
    2.2 The tag <TIMEX3>
           2.2.1 How to annotate TIMEX3s
           2.2.2 Attributes for TIMEX3
           2.2.3 BNF for the TIMEX3 tag
    2.3 The tag <DocCreationTime>
           2.3.1 How to annotate DocCreationTime's
           2.3.2 Attributes for DocCreationTime
    2.4 The tag <SIGNAL>
           2.4.1 How to annotate SIGNALs
           2.4.2 Attributes for SIGNAL
           2.4.3 BNF for the SIGNAL tag
    2.5 The tag <LINK>
           2.5.1 How to annotate LINKs
           2.5.2 Attributes for LINK
           2.5.3 BNF for the LINK tag
    2.6 Tags for Temporal Functions
           2.6.1 How to annotate Temporal Functions
           2.6.2 Attributes for Temporal Functions
           2.6.3 BNF for each Temporal Function tag
  3. Completely annotated examples

1. Introduction

This document describes the initial annotation guidelines for marking up text according to the TimeML language. It puts together all the modifications and additions discussed so far and gives us the tag specifications that follow for a first pass at TimeML.

For the sake of convenience, I&P(02) will be used to refer to Ingria and Pustejovsky (2002), and TIDES(02), to Ferro et al. (2002) throughout the whole document.

2. Explanation of TimeML Tags and their Attributes

2.1 The tag <EVENT>

We will consider events a cover term for situations that happen or occur. Events can be punctual (1-2) or last for a period of time (3-4):

  1. Ferdinand Magellan, a Portuguese explorer, first reached the islands in search of spices.
  2. A fresh flow of lava, gas and debris erupted there Saturday.
  3. Businesses are emerging on the Internet so quickly that no one, including government regulators, can keep track of them.
  4. 11,024 people, including local Aeta aborigines, were evacuated to 18 disaster relief centers.

We will consider also as events those predicates describing states or circumstances in which something obtains or holds true. However, they will be annotated only in cases where they are identifiably changed over the course of the document being marked up. For instance, in the following example, in the expression the Aeroflot Airbus the relationship indicating that the Airbus is run and operated by Aeroflot is not a State in the desired sense. Rather, because it is persistent throughout the event line of the document, we factor it out and it is not marked up. On the other hand, properties that are known to change during the events represented/reported in an article will be marked as States, as illustrated below. So, assume for now that events have only an id attribute:

All 75 people
<EVENT stid=1>
on board
</EVENT>
the Aeroflot Airbus
<EVENT eid=5>
died
</EVENT>

2.1.1 How to annotate EVENTs

Events are generally expressed by means of tensed and untensed verbs (1 and 2), nominalizations (3), adjectives (4), predicative clauses (5), or prepositional phrases (6):

  1. A fresh flow of lava, gas and debris erupted there Saturday.
  2. Prime Minister Benjamin Netanyahu called the prime minister of the Netherlands to thank him for thousands of gas masks his country has already contributed.
  3. Israel will ask the United States to delay a military strike against Iraq until the Jewish state is fully prepared for a possible Iraqi attack.
  4. ... after a Philippine volcano dormant for six centuries began exploding with searing gases, thick ash and deadly debris.
  5. "There is no reason why we would not be prepared," Mordechai told the Yediot Ahronot daily.
  6. All 75 people on board the Aeroflot Airbus died.
The annotation of formally simple events is straightforward (examples 1, 3 and 4 above): 
A fresh flow of lava, gas and debris 
<EVENT eid=1>
erupted
</EVENT>
there Saturday.
However, formally complex events may be sequentally discontinuous in some contexts:
  1. There is no reason why we would not be prepared.
    There is no reason why we would not be fully prepared.

  2. They will definitely take into consideration our readiness.
    They will definitely take it into consideration.
In order to avoid problems derived from tagging material of diverse scope, employ the following strategies:

2.1.2 Attributes for EVENT

a. Event ID number (eid)

Non-optional attribute. Each event has to be identified by an ID number. This will be automatically assigned by the annotation tool every time an EVENT tag is assigned to some string.

b. Class

Non-optional attribute. Each event belongs to one of the following classes:

c. Tense

Non-optional attribute. Given that the verbal tense is a valuable information in determining the temporal ordering of a related set of events, we will annotate it wherever it occurs. Indeed, it doesn't apply to nominalizations and non-finite clauses. In this case, the appropriate value will be 'NONE' (see below). The possible values for 'Tense' are:

d. Aspect

Optional attribute. Whenever possible, we will mark whether the aspect of the verb is perfective or progressive.

e. Polarity

Non-optional attribute. Two values are possible here:

2.1.3 BNF for the EVENT tag

attributes ::= eid class tense [aspect] polarity

eid ::= <integer>
class ::= 'OCCURRENCE' |'STATE' | 'REPORTING' | 'ASPECTUAL' | 'INTENDING'
tense ::= 'PAST' | 'PRESENT' | 'FUTURE' | 'IRREALIS' | 'NONE'
aspect ::= 'PROGRESSIVE' | 'PERFECTIVE'
polarity ::= 'POSITIVE' | 'NEGATIVE'

2.1.4 Complete annotation of some of the previous examples

  1. The young industry's rapid growth also is attracting regulators eager to police its many facets. 
    The young industry's rapid
<EVENT eid=1 class=OCCURRENCE tense=NONE polarity=POSITIVE> 
growth 
</EVENT>
also is 
<EVENT eid=2 class=OCCURRENCE tense=PRESENT aspect=PROGRESSIVE polarity=POSITIVE> 
attracting 
</EVENT>
regulators 
<EVENT eid=4 class=STATE polarity=POSITIVE>
eager 
</EVENT>
to
<EVENT eid=5 class=OCCURRENCE tense=IRREALIS polarity=POSITIVE> 
police 
</EVENT>
its many facets.
  2. Several pro-Iraq demonstrations have taken place in the last week. 
    Several pro-Iraq 
<EVENT eid=1 class=OCCURRENCE tense=NONE polarity=POSITIVE> 
demonstrations 
</EVENT>
have 
<EVENT eid=2 class=OCCURRENCE tense=PAST aspect=PERFECTIVE polarity=POSITIVE> 
taken 
</EVENT>
place in the last week.
  3. A fresh flow of lava, gas and debris erupted there Saturday.
  4. Prime Minister Benjamin Netanyahu called the prime minister of the Netherlands to thank him for thousands of gas masks his country has already contributed.
  5. Israel will ask the United States to delay a military strike against Iraq until the Jewish state is fully prepared for a possible Iraqi attack.
  6. ... after a Philippine volcano dormant for six centuries began exploding with searing gases, thick ash and deadly debris.
  7. "There is no reason why we would not be prepared," Mordechai told the Yediot Ahronot daily.
  8. All 75 people on board the Aeroflot Airbus died.
  9. 11,024 people, including local Aeta aborigines, were evacuated to 18 disaster relief centers set up at area schools.


2.2 The tag <TIMEX3>

Since the exact nature of the tag that TimeML uses to annotate temporal expressions is different in detail both from the TIMEX tag in STAG and the TIMEX2 tag in TIDES, we have decided to rename it TIMEX3. See I&P(02) for further details.

(WARNING: This section is subject to revision. Most of the specifications given here require common agreement. The section has been based on TIDES(02) treatment of temporal expressions. However, a deeper study of that document, and the corresponding tunning of the current section, is still required.)

2.2.1 How to annotate TIMEX3s

In order to be as much compliant as possible with TIDES TIMEX2 annotation, the TIMEX3 tag will be applied to TIMEX2 markable expressions. See TIDES(02) for detailed information about the particular expressions we intend to cover.

However, TimeML will differ from TIDES in the following issues (the examples given below are adapted from TIDES(02)):

  1. Extent of the time expression to tag (issue covered in TIDES(02), section 4.4.3).
    Contrary to the policy followed in the mentioned document, TIMEX3 will not include any of the following pre- and post-modifiers of the time expression:
  2. Treatment of complex temporal expressions (corresponding to section 4.4.4 in TIDES(02)).
    TIDES establishes different treatments for complex temporal expressions. In some cases, a unique tag is annotated. In some others, there is the possibility of embedding TIMEX2 tags. This is however not necessary in TimeML, given the availability of TEMPORAL FUNCTIONs and the LINK tag, which allow relating different time expressions. The treatment for complex temporal expressions in TimeML is as shown below:

Some examples of TIMEX3 markable expressions are the following:

  1. Dates and/or times:
    October 1, 1999
    9 a.m. Friday, October 1, 1999
    Fall 1999
    ten minutes to three
  2. Strict indexicals:
    now, nowadays, yesterday, currently, tomorrow, lately, etc.
    That is, tokens that occupy the entire value of VAL in TIDES guidelines (see TIDES(02), appendix B2), with the exception of the adjectives current, present, future, past, former, late, etc., which will be tagged as signals so that the appropriate temporal functions can be applied.

  3. All these kinds of expressions may appear in temporal expressions with Appositives (1), Range Expressions (2) or Conjoined Expressions (3):

    1. I'm a creature of the 1960s, the days of free love.
    2. She served as Canada’s ambassador to the U.N. from 1992 through 1995.
    3. The prime minister's visit is to run August 6-8.

    The annotation strategy that will be applied to them is exactly as detailed TIDES(02), p. 14.

At the present moment it is still pending a deep analysis of how much we want to benefit from TIDES annotation guidelines. All the specifications given above are provisional and require aproval.

2.2.2 Attributes for TIMEX3

a. Timex ID number (tid)

Each TIMEX3 expression has to be identified by an ID number. This is automatically assigned by the annotation tool.

b. Type

Non-optional attribute. Each event belongs to one of the following classes:

c. Calendar Date (calDate)

CalDate follows TIDES in using a modified version of the ISO 8601 time standard. The annotator should introduce the value assisted by the annotation tool.

If the calDate value is not given by the temporal expression (e.g., two days ago vs. October 1, 1999), the value will be computed by means of a TEMPORAL FUNCTION that will be designed also with the help of the annotation tool. This is deduced from the use of TEMPORAL FUNCTIONS in I&P(02).

2.2.3 BNF for the TIMEX3 tag

attributes ::= tid type calDate
tid ::= <integer>
type ::= 'DATE' | 'TIME' | 'COMPLEX'
calDate ::= PCDATA


2.3 The tag <DocCreationTime>

The DocCreationTime tag marks up the date of the article (DOA, in the STAG terminology). Since this is simply a temporal expression (TIMEX2 markable), it will be annotated as such. However, we will wrap it with the 'DocCreationTime' tag in order to preserve the distinguished status of this temporal expression as the primary temporal anchor of the article.

2.3.1 How to annotate DocCreationTime's

This is the only case in which we will use embedded tags: the date of the article has to be initially marked with the TIMEX3 tag, and then the whole structure wrapped up with the DocCreationTime tag: 

<DocCreationTime>
<TIMEX3 tid=1 type=DATE calDate=03271996>
03-27-96
</TIMEX3>
</DocCreationTime>

2.3.2 Attributes for DocCreationTime

DocCreationTime has no attributes.



2.4 The tag <SIGNAL>

A signal is a textual element that makes explicit the relation holding between two entities (timex and event, timex and timex, or event and event). Signals are generally:

2.4.1 How to annotate SIGNALs

Be it formally simple (e.g., before) or complex (no more than, end of, etc.), it will be annotated as a unit: 

<SIGNAL signalID=1>
no more than
</SIGNAL>
<TIMEX3>
two weeks
</TIMEX3>
However, we should be careful in not including two different signals under the same tag. This is particularly relevant in order to obtain a consistent application of temporal functions. Therefore, the sequence before last week must not be tagged like this: 
<SIGNAL sid=1>
before last
</SIGNAL>
<TIMEX3 tid=1 type=DATE>
week
</TIMEX3>
but like this: 
<SIGNAL sid=1>
before
</SIGNAL>
<SIGNAL sid=2>
last
</SIGNAL>
<TIMEX3 tid=1 type=DATE>
week
</TIMEX3>

In that example, last is the signal that helps in the temoral anchoring of week, whereas before relates the 'calDate' value that should hold for (last) week with another timex or event (cf. We ordered the ticket before last week.)

(N.B. The way how the 'calDate' value for (last) week is obtained will be discussed in section 2.6 on temporal functions.)

2.4.2 Attributes for SIGNAL

By now SIGNAL has only one attribute: 'signalID'. It is automatically assigned by the annotation tool each time a SIGNAL is marked up.

2.4.3 BNF for the SIGNAL tag

attributes ::= sid

sid ::= <integer>


2.5 The tag <LINK>

2.5.1 How to annotate LINKs

2.5.2 Attributes for LINK

2.5.3 BNF for the LINK tag

attributes ::= (eventID | timeID) [signalID] (relatedToEvent |
relatedToTime) relType magnitude

eventID ::= <integer>
timeID ::= <integer>
signalID ::= <integer>
relatedToEvent ::= <integer>
relatedToTime ::= <integer>
*relType ::= 'BEFORE' | 'AFTER' | 'INCLUDES' | 'IS_INCLUDED' |
'SIMULTANEOUS' | 'IAFTER' | 'IBEFORE' | 'ID' | 'INITIATES' |
'CULMINATES' | 'TERMINATES' | 'CONTINUES'
magnitude ::= <integer>


2.6 Tags for Temporal Functions

2.6.1 How to annotate Temporal Functions

2.6.2 Attributes for Temporal Functions

2.6.3 BNF for each Temporal Function tag



3. Completely annotated examples