The UNL documentation available from the UNL website at http://www.undl.org/unlsys is written in a rather technical and detailed fashion all at one level which makes it quite difficult to obtain a general understanding of the principles involved without lengthy detailed study. Some of it is very complicated and there is little or no explanation of why things are the way they are.
 
I shall try to describe the system of Universal Words in a top down fashion by discussing general principles first though this will inevitably be at the expense of some (important) details which I hope to be able to deal with later or through discussion subsequently on this e-mail list. It may be a help to have a look at the list of universal words which can be seen at http://www.undl.org/unlsys/uw/UNLKB_tree.txt.
 
Please comment on this list with criticism or for clarification.
 
The Universal Word (UW)
 
 
1. Introduction
 
By the very nature of an interlingua each word in its dictionary must represent only one meaning.
 
A source language must be able to choose the appropriate interlingual word for a word in its own language and a target language must be able to know what that interlingual word means in order to be able to choose the appropriate word in its own language so as to be able to effect an accurate translation.
 
The UNL has only the four primary types of words, namely Nouns, Verbs, Adjectives and Adverbs. Other categories, such as Prepositions, Conjunctions, and Comparatives etc. are either ignored as being language specific or are indicated by other mechanisms.
 
The principal file for storing word meanings is the Master Definition file held and maintained by the UNL organisation. Each Language Server holds its own dictionaries with words in its own natural language corresponding to words in the central Master Definition file.
 
The Master Definition file is never finalised and is being added to all the time. I think individual language developers propose the words they need when those words are not already in the Master Definition file.
 
At first sight it may be rather unexpected to find that the prime movers in the development of UNL (Japanese) in the UN University have chosen English(American) words as the basis of their interlingua. However, this does not mean that the interlingua is the English language. I hope the following information will show how what one might call the normal meanings of a word can be individually restricted to single unique meanings for the purposes of this interlingua.
 
 
2. Restraints
 
It is known only too well that individual words practically always have many different meanings depending on context, either through syntax, that is, what parts of speech they represent, and through semantics, that is, what they mean in the context.
 
It is also well known that words hardly ever translate between languages on a one to one basis so an interlingua needs somehow to indicate specific meanings for which it should be easier to find suitable matching words in other languages.
 
The way to indicate a particular meaning of a word in UNL is to attach what are called "restraints" sufficient to restrict the meaning of the word uniquely. For example "swallow(icl>thing)" would indicate that this "swallow" was a "thing" and was therefore a "nominal concept" and not, say, a verbal concept, in which case "swallow(icl>do)" could be used.
 
The words like "thing and "do" here have their normal meanings in English but are defined elsewhere in the Master Definition file.
 
The relation label, "icl", is used to indicate which more general meaning the word is included in. There are many others, such as "iof" which indicates what thing the word is an instance of, "pof", which indicates what thing the word is part of, etc. some of which are discussed further below.
 
 
3. Semantic space
 
In the previous section we saw the way in which UNL can indicate that the meaning of a word can be said to fall within the meaning of a more general word. In other words, each word can be said to have a meaning space which can be divided up into "smaller" units. Each universal word still has its own normal meanings in English but the specification of one within any particular semantic space should indicate a particular unique meaning.
 
In fact, UNL's semantic space starts with the concept of a Universal Word, or UW, which contains all possible meaning. This is subsequently divided up as follows (rightly or wrongly, I leave you to judge) into four main groups.
 
nominal concept(icl>uw)
verbal concept(icl>uw)
adjectival concept(icl>uw)
adverbial concept(icl>uw)
 
Some lower levels are,
 
thing(icl>nominal concept>uw)
 
abstract thing(icl>thing>nominal concept>uw)
concrete thing(icl>thing>nominal concept>uw)
functional thing(icl>thing>nominal concept>uw)
place(icl>thing>nominal concept>uw)
volitional thing(icl>thing>nominal concept>uw)
 
do(icl>verbal concept>uw) refers to something done
occur(icl>verbal concept>uw) refers to something that happens
be(icl>verbal concept>uw) refers to something that is
 
 
4. Nominal concepts
 
At this point, having expanded "nominal concept" and "verbal concept" here are some examples of "nominal" universal words.
 
mind(icl>abstract thing)
matter(icl>concrete thing)
food(icl>functional thing)
space(icl>place)
being(icl>volitional thing)
 
These words have other meanings but with the restraints in the Master Definitions file they point to particular unique meanings.
 
Note that it is not necessary to include ">thing>nominal concept>uw" each time as these restraints are already declared for "abstract thing", "concrete thing" etc. elsewhere in the dictionary. What matters is that "mind(icl>abstract thing)", "matter(icl>concrete thing)", etc. are unique entries in the dictionary.
 
I think these two principles apply to all universal words.
 
 
5. Verbal concepts
 
Here are some examples of "verbal" universal words.
 
read(icl>do(agt>thing,obj>thing))
read(icl>do(agt>person,obj>information))
 
Here we have examples of multiple restraints and there can be as many as necessary. "agt" refers to the agent, or the initiator of an action, and "obj" refers to the object, or the direct recipient of an action.
 
Presumably, in this case, in at least one language different words are used for each of these two definitions of the verb "read". In practice the words "icl>do" are actually left out as the two entries are still different and both unique within the dictionary by virtue of the remaining restraints.
 
In the next example "gol" refers to the goal, or the result, of an event (In the UNL documentation an "event" appears to refer to both actions and occurrences).
 
melt(icl>occur(gol>thing,obj>thing))
melt(icl>occur(obj>thing))
 
Two versions for occurrencies of melting are catered for, one with the result of melting specified. Again the words "icl>occur" are in practice actually left out as the two entries would still be different (Personally I wonder whether this practice is safe since in the future we could have the action of deliberate melting requiring "icl>do" but I have to admit that this would probably also require us to specify an agent, eg. "agt>person", in the restraints, which would make the UWs different again).
 
 
6. Conclusion
 
I hope this introductory information which I have gleaned from the UNL documentation is all correct. In any case, if you have any comments about it please mail them to this list.
 
I have not yet discussed adjectival and adverbial concepts as this message is already too long and the way these two concepts are handled in UNL are rather more difficult to understand. In fact, I am still having problems with them!