MIKROKOSMOS

Creating Mikrokosmos TMRs

Text Meaning Representation Frames for Mikrokosmos

Introduction

This tutorial assists in the translation of a text in a natural language into the text meaning representation used by Mikrokosmos.

This tutorial offers a step-by-step analysis of a sample text in English and explains how the various slots and fillers available in the Mikrokosmos interlingua transform the natural language text into an interlingua text.

This tutorial will teach you how to:

Prepare the system and the text for translation
Separate sentences of the text into appropriate clauses
Determine a head for each clause
Assign case and circumstantial roles for each head
Define types of textual and non-textual relations for each head as necessary
Use required notation accurately

Translating the Text

If the text is in a language other than English, it must be translated into English before beginning the translation into interlingua.

When the text is in English, you are ready to begin.

Setting up the TMR Shell

[Need information from Ralf on how to set up a TMR shell]

The TMR shell will look like this:

[include a picture or representation of the TMR shell]

As you can see, the TMR shell provides all possible slots so that you will only need to provide fillers. Those slots which do not apply to the text fragment you are transcribing may be left blank.

Working Clause by Clause

It is easiest to begin transcribing your text into ILT if you first separate the sentences of your text into clauses. A clause, generally speaking, is a subject and finite verb (a verb which changes tense with the subject). So there may be one, or several, clauses in a sentence.

Let's look at an example. Consider the following text:

Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation, to set up a joint venture in Brazil possibly this summer to produce ferrosilicon, reported Nikkan Kogyo.

To separate this text into clauses, first highlight or pull out the verbs or verb phrases of each sentence:

Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation, to set up a joint venture in Brazil possibly this summer to produce ferrosilicon, reported Nikkan Kogyo.

so we have our clauses:

Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation
to set up a joint venture in Brazil possibly this summer
to produce ferrosilicon
reported Nikkan Kogyo

Notice that although this text is only one sentence long, there are at least four clauses in it. Also notice that when you have a verb, a subject is implicit, even if it is not mentioned explicitly in the text (an action infers someone or something is doing the action).

It can be overwhelming to attempt to transcribe an entire sentence, so it is important at this stage to only transcribe one clause at a time. So first, transcription of the first clause.

First Clause:

Selecting the Head

A verb is a common head in a TMR frame. In the case of our example, the verb phrase has reached agreement presents itself as a possible head. However, it would be a mistake to use the phrase "as is" for two reasons:

Has reached is the present perfective (?LOOK UP) of the verb to reach, and verb tense should never be put into the head of a frame.
To reach agreement is a stylized verb phrase which basically means agree. While there are other shades of meaning which to reach agreement captures which are not reflected in agree, these other meanings should be expressed later in another slot of the TMR frame we are currently writing.

Consequently, agree is the verb we should use as the head of the current clause.

Filling the Slots for Case Roles

The next step to representing the first clause is to fill the "case roles, which are the arguments a predicate can take. These include:

Clicking on any of these terms will take you to a definition and examples of that term.

In addition, the circumstantial roles must be filled in if they are appropriate to the clause. These are roles which relate events to more circumstantial pieces of information that describe them, such as:

location
time
Clicking on any of these terms will take you to a definition and/or examples of that term.
But what about the clause we're working on? To refresh your memory, here it is again:

Kawasaki Steel Corp. has reached agreement [our head: agree] with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation

Going down the list of available case roles, let's begin by seeing if agree has an agent.

Agent
To find out if agree has an agent, ask "who or what is agreeing?" The answer to that in this case is Kawasaki Steel Corp., Mitsubishi Corp. and CIA Vale Do Rio Doce. So we have three "agents" of agree.
If only things were that simple! But there is more to the text than that. Note that in the original clause, "Kawasaki Steel Corp." is given special mention; it has "reached agreement" with the other two companies, which are placed in a clause-end prepositional phrase. So, in fact, "Kawasaki Steel Corp." is the "agent," while the other two companies fill the case slot accompanier. The
accompanier slot in this way conveys the idea that the emphasis in this clause is placed on Kawasaki Steel Corp.
However, we immediately run into a problem: each slot can take only one filler, so how can we put two companies into one slot?. To handle multiple fillers in a slot, we use a set notation:
```
;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with
;; Mitsubishi Corp. and CIA  Vale Do Rio Doce

%agree_1
	agent		%company_1
	accompanier     %set_1

%company_1
	name		"Kawasaki Steel Corp."
```
The accompanier slot has been filled, but now % set_1 must be defined (as we have defined the agent slot).
Set Notation
As used by Mikrokosmos, set notation is a vastly versatile tool for defining all types of groups and subgroups. In the present case, we need only use a few of the basic slots available to set notation. First, we need to specify what type of members the set has. The fillers for the accompanier slot are companies, so we need only the ontological concept company. Also, there are two of them, so we simply invoke the cardinality slot as well.
```
;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with
;; Mitsubishi Corp. and CIA  Vale Do Rio Doce

%agree_1
	agent		%company_1
	accompanier	%set_1
	            		member-type	  *company
		    		cardinality 	  2
```
Since we know the members of the set, we can also use the slot elements, which is used to name a set's members:
```
;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with
;; Mitsubishi Corp. and CIA  Vale Do Rio Doce

%agree_1
	agent		%company_1
	accompanier	%set_1
	            		member-type	  *company
		    		cardinality 	  2
				elements      (set "Mitsubishi Corp"
                                       		   "Cia Vale Do Rio Doce")
```
Note that we have nested another set inside set_1; this is perfectly acceptable with Mikrokosmos' notation.

Theme
A very common pairing of case roles finds agent and theme occurring together.
To discover if agree requires a theme case role, ask yourself: what does the action of agreeing affect? Often a theme will be an entity of some kind, but in this case, agree affects the next clause, that headed by create:

Kawasaki Steel Corp. has reached agreement [agreed] [...] to set up a joint venture [...]
So it is the action of setting up which is affected by agree. This is a very common use of the theme slot--filling it with an action that is "nested" within another action:
```
;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with
;; Mitsubishi Corp. and CIA  Vale Do Rio Doce

%agree_1
	agent		%company_1
	accompanier	%set_1
	            		member-type	  *company
		    		cardinality 	  2
				elements      (set "Mitsubishi Corp"
                                       		   "Cia Vale Do Rio Doce")
        theme   %set-up_1
```
There are no other case roles which need to be invoked for the first clause, but there are circumstantial roles which are part of every predicate frame.
Finishing the Predicate Frame
Each head has time and aspectual slots because every action or state happens in the realm of time, in relation to other actions or states; in addition, since a text presupposes a speaker who bears an attitude toward what s/he is saying, often an attitude becomes part of what needs to be represented in a TMR. In the section of text we are looking at, attitude doesn't seem to be present as a component of the text. In the frame for agree, then, the ontological concepts of the time and aspectual slots is instantiated:
```
;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with
;; Mitsubishi Corp. and CIA  Vale Do Rio Doce

%agree_1
	agent		%company_1
	accompanier	%set_1
	            		member-type	  *company
		    		cardinality 	  2
				elements      (set "Mitsubishi Corp"
                                       		   "Cia Vale Do Rio Doce")
        theme         %set-up_1
        time          %time_1
        aspect        %aspect_1
```
The task then is to define each of these fillers separately.

Time
The filler for the time slot,
```
%time_1
```
marks that agree happened at a particular time, which will be discussed more elsewhere. For the time being, it is enough to know that it acts as a marker which allows the time at which agree happened to be referred to.

Aspect
The aspect slot can be filled by any or all of the slots phase, iteration, duration and telicity. It represents the aspect of the head of the frame--in this case, agree.
There are several possible aspectual slot fillers which may be used.
In the case of agree, phase would be filled by end because of the phrase reached agreement in the original text. Iteration would be single since the act of agreement happened only once, duration would be momentary, and the telicity slot would be filled by "true," since the act of agreeing is complete:
```
;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with
;; Mitsubishi Corp. and CIA  Vale Do Rio Doce

aspect_2
	phase		end
	iteration	single
	duration	momentary
	telicity	true
```
The nearly complete frame then has become:

;; gloss: Kawasaki Steel Corp. has reached agreement [agreed] with ;; Mitsubishi Corp. and CIA Vale Do Rio Doce %agree_1 agent %company_1 accompanier %set_1 member-type *company cardinality 2 elements (set "Mitsubishi Corp" "Cia Vale Do Rio Doce") theme %set-up_1 time %time_1 aspect %aspect_1 %time_1 absolute-time {recent past} %aspect_1 phase end iteration single duration momentary telicity yes

Property Frames
There is one portion of the text in the first clause which has been left wholly unrepresented:

Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation [...]

The only "verb-like" word in this text segment is state-run; one option might be to decompose the word into the verb run. However, there is a better way to represent this text than to decompose the verb-like phrase into a head of its own.
Because state-run and mining are properties, they more approriately belong to a property frame. A property frame is headed by an object, and has special slots suited to describing common object properties.
Because Mikrokosmos deals primarily with texts from the business domain, it is quite common to come across parenthetical information such as this description of a corporation. For that reason, special slots have been developed for this purpose. They are:

activity/service
product
ownership
corporate relations
Examples of company slot fillers will make their use more clear. A thorough listing of company slot fillers is also available.
To begin our property frame, we must first instantiate the company we are describing by giving it a name. We instantiate it by invoking the ontological concept company and by marking it with an arbitrary number. To discover if there is an ontological concept available which will describe the concept we want to instantiate, we [describe what is done with Mikrokarat to search ontol. db].
We discover there is an ontological concept called company, so we head the frame with this. We would also use the activity/service slot and the ownership slot to represent state-run mining corporation.
;; Brazil's state-run mining corporation %company_1 name "Cia Vale Do Rio Doce" activity *mining ownership $Brasil
The symbols used in this property frame have special uses and procedures for how they should be used.
The first clause frame is now complete.
Second Clause:

Dividing the Clauses Correctly
We initially divided the second clause as:
...to set up a joint venture in Brazil possibly this summer...
However, if we look ahead to the next clause:
...to produce ferrosilicon...
we may want to ask ourselves, is possibly this summer meant to modify set up a joint venture or to produce ferrosilicon? One may make arguments for either choice--welcome to the wonderful world of text meaning representation!
We will take the position that it is the setting up of the joint venture which will take place possibly this summer instead of the producing of ferrosilicon. This will require using the attitude slot, which we will do when we finish the frame for this clause.

Filling the slots for Case Roles
Because we needed to put the head of the second clause in as filler for the theme slot of the previous clause, we already know that the head of the second clause is set up. It can be hyphenated:

;; to set up a joint venture in Brazil ;; possibly this summer %set-up_1
It is only necessary to look again at the definitions and examples of each to determine which case roles are needed.

Agent
Although the agent doesn't occur in the second clause, it can be traced back to the first:

First Clause
Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation...

Second Clause
...to set up a joint venture in Brazil...
Asking, "who or what is setting up something?", we get the answer--Kawasaki Steel Corp., Mitsubishi Corp. and CIA Vale Do Rio Doce. So we instantiate another set:

;; gloss: to set up a joint venture in Brazil ;; possibly this summer %set-up_1 agent %set_2 member-type *company cardinality 3 elements (set "Kawasaki Steel Corp." "Mitsubishi Corp" "Cia Vale Do Rio Doce")

Theme
The entity whose state is affected by the action of setting-up is clearly joint venture.
However, there is more to the "joint venture issue" than meets the eye. Corpus analysis was conducted, and it was decided that because the meanings of the term are multifarious in Japanese as well as English, the best representation of the term would be in the form of a predicate head create and a clearly non-language-specific theme tie-up. Since setting-up a joint venture can be glossed as creating a tie-up, we can double back and change the head to create. (This also necessitates changing the theme of the first clause to create, which is easily done.)
The second clause thus far becomes:

;; gloss: to set-up [create] a joint venture [tie-up] in Brazil ;; possibly this summer %create_1 agent %set_2 member-type *company cardinality 3 elements (set "Kawasaki Steel Corp." "Mitsubishi Corp" "Cia Vale Do Rio Doce") theme %tie-up_1

Finishing the Predicate Frame: Circumstantial Roles
Are there any other roles which should be represented in this clause of the TMR? Look at the circumstantial roles and their definitions to discover if there are any appropriate to represent this clause.

Location
The circumstantial role location probably caught your eye; a prepositional phrase such as in Brazil is a giveaway that the location slot is needed:

;; gloss: to set-up [create] a joint venture [tie-up] in Brazil ;; possibly this summer %create_1 agent %set_2 member-type *company cardinality 3 elements (set "Kawasaki Steel Corp." "Mitsubishi Corp" "Cia Vale Do Rio Doce") theme %tie-up_1 location $Brazil

Purpose
To find out if there are additional roles, it is often necessary to look ahead to the clauses which follow. They are:

Third Clause
to produce ferrosilicon

Fourth Clause
reported Nikkan Kogyo

It is clear from the text that the purpose of creating the tie-up is in the third clause: to produce ferrosilison; in fact, the insertion of the words "in order to" make this relation clear--to set-up [create] a joint venture [tie-up] in Brazil possibly this summer [in order to] produce ferrosilicon. So the purpose slot should be added, with the head of the next clause as its filler:

;; gloss: to set-up [create] a joint venture [tie-up] in Brazil ;; possibly this summer %create_1 agent %set_2 member-type *company cardinality 3 elements (set "Kawasaki Steel Corp." "Mitsubishi Corp" "Cia Vale Do Rio Doce") theme %tie-up_1 purpose %produce_1 location $Brazil
The aspect, time and attitude slots remain:

;; gloss: to set-up [create] a joint venture [tie-up] in Brazil ;; possibly this summer %create_1 agent %set_2 member-type *company cardinality 3 elements (set "Kawasaki Steel Corp." "Mitsubishi Corp" "Cia Vale Do Rio Doce") theme %tie-up_1 purpose %produce_1 location $Brazil aspect %aspect_2 time %time_2

Aspect and Time Slots
Since create is an achievement, its phase can be described as end; it has a single iteration, is of momentary duration, and it is a complete action (telicity)>:

%aspect_2 phase end iteration single duration momentary telicity yes
The slot time, again, marks that create happened at a particular time which can now be referred to, and which will be discussed more elsewhere:

%time_2 absolute-time {future}

Attitude
This clause does need an attitude slot; the word "possibly" is the trigger that invokes the slot. The attitude slot is where emotion and feeling present in the text is transcribed, either of the "speaker" of the text or of a participant in the text.
In this case, "possibly" reflects the belief of the speaker of the text that the action of %produce_1 "this summer" is only possible--not a sure thing, but potentially could happen. There is a filler for the attitude slot called potential which can express "possibly":

POTENTIAL: The scale of the potential attitude goes from "X is not possible" (value 0) through "X is possible" (value 1.0).

The slots to be filled for attitude follow; we have filled in the type slot with potential; the value, as indicated in the definition above, is "1":

%attitude_4 type potential value 1 scope attribution
The scope slot is filled with the slots or heads which are encompassed by the attitude. In this case, the potential attitude scopes over %produce_1. The attribution slot is filled by the speaker or other text participant who possesses the attitude. In this case, it is the three companies who possess the attitude, so we create another set, which we can cross reference with %set_2 (which has the three companies as its members). The finished attitude slot then is:

attitude_1 type potential value 1 scope %produce_1 attribution %set_3 ;; cross-referenced with %set_2
The second clause frame is now complete.
Third Clause:

Filling the slots for Case Roles
Because we needed to put the head of the second clause in as filler for the theme slot of the previous clause, we already know what the head of the second clause is:

;; gloss: to produce ferrosilicon %produce_1
Although it may seem that the agent of produce would be the same set as in the previous clause, a closer look is in order. The previous clause, joined to our third clause, yields:

to set up [create] a joint venture [tie-up]
possibly this summer to produce ferrosilicon
So we see that it is the newly-created joint venture which will produce ferrosilicon. We must instantiate a new "tie-up," and coreference it with %tie-up_1:

;; to produce ferrosilicon %produce_1 agent %tie-up_2 ;; coreferenced with %tie-up_1

Theme
This time, instead of filling the theme slot with another head, we will fill it with an object-- ferrosilicon:

;; gloss: to produce ferrosilicon %produce_1 agent %set_3 ;; coreferenced with %set_2 theme %ferrosilicon_1

Finishing the Predicate Frame
As always, representation of the head must always include aspect and time information. This time, these circumstantial roles are filled with less typical information.

Aspect
The slots for aspect this time are filled with information modified by the words surrounding the head, in particular, this summer. Because "production" will be going all summer, aspect will be as follows:
%aspect_3 phase continuing iteration multiple duration prolonged telicity no
Note the "continuing" phase and the "no" filler of the telicity slot, indicating that the action is not complete.

Time
Because of the phrase this summer, the time slot is defined differently. One should discover the date of the text and extrapolate the dates within which the act of production should operate. In this case, say the text was dated January 14, 1986. The time enterer into the time slot should be:

%time_3 absolute-time {010686-010986}
That is, the first day of June, 1986, through the first day of September, 1986.
The third clause frame is now complete.
Fourth Clause:

"Reporting" Verbs
The final clause is a speech act clause which is reporting what preceded in the sentence:

Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation, to set up a joint venture in Brazil possibly this summer to produce ferrosilicon, reported Nikkan Kogyo.

This is similar to the speech acts one finds in novels ("I told you not to come," said Pierre.") and even in everyday conversation (My dad said he's going to to kill me if I wreck the car again.). Of course, they are very often found in newspaper articles, which is where our sentence comes from.
By now, you should be able to tell what's coming:

;; gloss: reported Nikkan Kogyo %report_1 agent %company_2
In turn, %company_2 is defined as:

%company_2 name "Nikkan Kogyo"
The theme, in this case, is the first predicate of the sentence; imagine that the fourth clause was at the beginning of the sentence--Nikkan Kogyo reported that Kawasaki Steel Corp. has reached agreement with... So we insert into the theme slot the first head of the sentence, agree:
;; gloss: reported Nikkan Kogyo %report_1 agent %company_2 theme %agree_1
Next, we fill our time and aspect slots (the attitude slot is not needed):

time %time_4 aspect %aspect_4
and define them:

%time_4 absolute-time {recent past} %aspect_4 phase end iteration single duration momentary telicity yes
The fourth clause frame is now complete.
Time Relations
Time relations are given at the end of the TMR; there is no reference to time relations within the TMR body. A time relation indicates if an event occurred at, after, or during another event. Determining time relations between all of the events of a TMR can be a difficult job; the typical TMR may have several complex sentences and may need as many as 15 or 20 time relations (or more). In the current case, we have only five heads which require a time relation slot.

Using the Tools
[Insert text on how to do temp rels with the tools]
Arranging Events
The easiest way to decide how to relate the heads of each frame to each other is to arrange them around the time of the publication of the article, which is traditionally at %time_0.
The heads once again are:

Kawasaki Steel Corp. has reached agreement with Mitsubishi Corp. and CIA Vale Do Rio Doce, Brazil's state-run mining corporation

to set up a joint venture in Brazil possibly this summer

to produce ferrosilicon

reported Nikkan Kogyo

A little figuring gives us:

the date of publication took place after reporting (%time_0);
reporting took place after agreeing
setting up [creating] a joint venture [tie-up] will take place after reporting;
producing will take place after creating.

Notation for Time Relations
Now we will see why it was important to mark each of the clause frames with a %time_n.
A little chronological rearranging gives us the following order of events:

agree (%time_1)
report (%time_4)
date of publication (%time_0)
set-up [create] (%time_2)
produce (%time_3)
[Explanation here of how tools make this oh so easy...] The following example shows notation for time relations:

%temp-rel_n type after arg-1 %time_1 arg-2 %time_2
where type could be at, during, or after.
Let's begin with agree:

;; gloss: "report" takes place after "agree." %temp-rel_1 type after arg-1 %time_4 arg-2 %time_1
Moving down the list, we have:

;; gloss: date of publication takes place after "report." %temp-rel_1 type after arg-1 %time_0 arg-2 %time_4 ;; gloss: "set-up" [create] takes place after date of publication. %temp-rel_1 type after arg-1 %time_2 arg-2 %time_0 ;; gloss: "produce" takes place after "set-up" [create]. %temp-rel_1 type after arg-1 %time_3 arg-2 %time_2
As you can see, once you have arranged the events in "regular English," plugging in the appropriate values and arg fillers is much easier.
Table of Contents:

Using the Tools
[Since I expect that this will be a tool-intensive job because of its mechanical nature, this section will be written later.]

Author: Donalee H. Attardo (dattard@crl.nmsu.edu)

MIKROKOSMOS

Creating Mikrokosmos TMRs

Introduction

Translating the Text

Setting up the TMR Shell

Working Clause by Clause

Selecting the Head

Filling the Slots for Case Roles

Agent

Set Notation

Theme

Finishing the Predicate Frame

Time

Aspect

Property Frames

Dividing the Clauses Correctly

Filling the slots for Case Roles

Agent

Theme

Finishing the Predicate Frame: Circumstantial Roles

Location

Purpose

Aspect and Time Slots

Attitude

Filling the slots for Case Roles

Theme

Finishing the Predicate Frame

Aspect

Time

"Reporting" Verbs

Using the Tools

Arranging Events

Notation for Time Relations

Using the Tools