Practical How-to Guide using AWB for Metonymy Tagging
Guide for LDC Annotators

Set up your current tagging session

1) open an xterm

2) type awb& to start the interface

3) open the file you wish to tag by

pull down FILE menu
select Open Document (Latin-1)
dialog box will open
go to Directory: pulldown menu and select file from the appropriate directory
 (follow this link to get an overview of where the data is stored)
click on filename and click Open
4) dialog box will open
 select      *Source file contains SGML
                 *No normalization
 click  Do It
5) 'Tagset Contents' dialog box will open
 click  Load File
6) you should now see a single story displayed in the interface

7) After the file has been loaded, load in the ACE annotation tag
  preferences file:

pull down OPTIONS menu
select Tag Preferences
select Load Tag Preferences
select mention-prefs-for-edt+
The system will then remove the old tagset and load the ACE tagset.

8) After the mention-prefs have been loaded, load in the
  "PilotEntity.rtd" relations definition for tagging relations:

pull down OPTIONS menu
select Relations
select Edit Existing Relations
This step should open up a new window displaying the relations table for this file.

Make sure the pilotentity box under OPTIONS -> Relations is selected to see the table.]

9) Go to OPTIONS -> File Saving Options
Be sure that Generate ACE Pilot Format is NOT selected
Click OK

You should only need to do this once ever, but please double-check the first few times you run the program.

10) Now you're ready to annotate the file.


Metonymy Annotation

1) Some basics
M1, M2, M3 = Mouse button 1 (Leftmost), 2 (Middle), 3 (Rightmost)
To select words, click on them with M1.  Clicking on M1 or M3 will select words to the left or right of the first word selected.
Hover = place cursor over a word but don't select it
Cycle = place cursor over a tagged word, hit TAB to see multiple layers of annotation on that word/set of words

2) Annotating Classic Metonymies
Here your goal is to create two mentions of the same string of text, one of which will be tagged with the mention's "literal" type and one the "intended" type.

To add the LITERAL mention type:
a) Hover over the text you wish to annotate.
b) Cycle through annotations until you get the GREEN layer to show up.  For a multi-word phrase, you may need to hover over the space between words.
c) With the GREEN layer on top, call up the Edit Annotation window by hitting Meta-Control-M1
The first column of this window should contain tags for RELSLOT, RELNAME, ANNOTID.  The window should look similar to this:
d) Select the desired REF (Literal) and hit Add.
[d') Check the original TYPE of this mention in the relations table.  If it is GSP, then you also must add the correct ROLE for the GSP.]
e) You should now notice that the font color has changed for the words you've tagged in the text..

To add the INTENDED mention type:
a) Select the text with M1.
b) Add the text you've selected into the appropriate cell in the Relation Table using M2. (It may need to go in a new row by itself, or get added as a mention of a pre-existing entity.)
c) Now that you've added another mention to the relation table, you can add the Intended type to this mention by going to the TYPE column in the relation table and selecting the correct TYPE with M3.
d) Now you must add the correct REF (Intended) for this type.  Go back to the text window, and hover over the mention (it should appear in GREEN highlighting with black font), then calling up the Edit Annotation window by hitting Meta-Control-M1.  The window should look like this, and should NOT have any pre-existing ROLE/REF annotations in it.

Select the desired REF (Intended) and hit Add.
[d') If the TYPE of this new Intended mention is GSP, then you also must add the correct ROLE for the GSP.]

After this process is complete, check your annotations by seeing what appears at the bottom of the text file when you hover over the text.  Cycle through the levels of annoation and check the text at the bottom.  You can tell which mention is on top by looking at the color of the font type.  LITERAL is white; INTENDED is red.

2) Annotating Metonymic Composites (GPEs)

This process is similar to that of annotating classic metonymies, but a little simpler.
a) hover over the word or phrase you wish to add
b) If necessary, cycle through the existing tags until the GREEN highlighting is on top.  For a multi-word phrase, hover over a word that is not the head.
c) Call up the Edit Annotation window by hitting Meta-Control-M1.  The window that appears should look like this:

d) Add a ROLE for the GPE, then click Add.  In the text window, the mention you've just annotated will have a different color font.

After this process is complete, check your annotations by seeing what appears at the bottom of the text file when you hover over the text.  Cycle through the levels of annoation and check the text at the bottom. You should see a tag that looks something like <RelationSlotString-XXX-Role> where XXX is the specific role (ORG, GPE, PER, LOC) that you just entered.

--------------
Some features to help you perform Metonymy Annotation
(from Lisa Ferro)

Finding Mentions in the text and table

Control-button-3 on an entry in the table and the text widget will
scroll AND highlight the string in the text.  This is also now the way
to see the mention and head at the bottom of the PilotEntity table.
(Merely "hovering" doesn't work anymore. Marcia says she can put the
hover-ability back in if we want it.)  To undo the highlighting and
display caused by using control-button 3, you use
control-double-button-3 on the entry in the table.

Control-button-3 on a tagged item in the text widget and it will cause
the corresponding entry in the table to be highlighted. (Unfortunately
we don't know how to make the table widget scroll to where
the highlighted entry is, but we're working on it.) To undo the
highlighting you have just caused by using control-button 3, you use
control-double-button-3 on the string in the text.

TAG STACK ORDER

You'll notice that the MENTION tags default to
being on top of the green tags IF they have the same extent.  If you
hover your cursor over the tag and press the Tab key, the green tag will
rise to the top.  At first I asked David to fix it so the green tags
would default to being on top, but then I discovered something...  One
of the things we're supposed to do is check the existing annotation.
When there are long strings with multiple embedded tags, if the green
tags are on top, you just see one long green tag, and can't tell where
the embedded tags begin and end.  I found that it was useful to be able
to "raise" the embedded green tags one constituent at a time, allowing
me to see the extent of each one.

However, there are admittedly two disadvantages to having the MENTION
tags default to being on top:

(a) You have to Tab on *each* MENTION tag (whether or not it's
embedded)
simply to assure yourself that there really is a green tag for that
span.

(b)  The table-to-text highlighting is hidden if there is MENTION tag
over the green tag.  The screen will still scroll so that the string is
visible in the text window, but you'll have to eyeball the text to find
it. To help make this easier, we've just this minute added an underline
to the hightlighted string. So that should help you locate the string,
even if you can't see the highlighting.
 

A note on annotating "true metonyms":

The ppt file I sent you doesn't tell you
what to do about strings that require MENTION/MENTIONHEAD tags.  The
answer is that if a string invokes two entities, you only need to give
it two table entries.  You do NOT need two sets of MENTION/MENTIONHEAD
tags. However, the same warning  I sent you last week about
deleting/editing tags now applies doubly:  If you want to change the
MENTION/MENTIONHEAD tags, you will have to delete and re-do the
green tags as well.  If it's a "true metonym," you'll have two green tags to
re-do.
 



strassel@ldc.upenn.edu
8/15/2001