>Trying to produce our latest milestone report we argued about the
>correct usage of the dash '-' to combine words. 

What you're talking about is generally referred to as a "hyphen",
rather than a dash.  Hyphen is an ASCII character (#45 decimal), and
appears on QWERTY keyboards to the right of zero; typically it's also
used for the minus sign and for numerous textual separation duties.
Whereas a dash -- frequently typed as two --- or even three ---
successive hyphens -- is used like a parenthesis.  See Lewis Thomas's
classic "On Punctuation" for further details about the dash, which is
more of a typesetter's punctuation than a typist's.

>The question was when to use it and what semantic changes it implies.
>An example would be 'user-interface' vs. 'user interface' or
>'user-model' vs. 'user model'.

Hyphenation rarely has any semantic effect, and even then it's
arbitrary, not regular. It's similar to the problem of whether to
hyphenate German compound nouns. The unmarked case in German is to
write them without spaces, and in English to separate them with
spaces.

But in both cases, some are hyphenated.  Which ones is rarely a matter
of logic; these conventions (like ASCII itself) come about arbitrarily
and rarely have any advantage other than one's own familiarity.  When
different people find themselves feeling familiar with different
versions, you get variation.  But no regular semantics.

>My working hypothesis is that the dash implies that the first word
>modifies the second, whereas without dash the second modifies the first.

Intriguing hypothesis, that.  Very nice and regular.  Unfortunately,
there's as much evidence for it as against it.  Plus, "modification"
doesn't help you in pinning it down.  It's full of largely arbitrary
idioms, too.  Consider, as a small example, that "Bill's picture" can
mean
      the picture that depicts Bill (somehow)
      the picture that Bill owns
      the picture that Bill made (somehow)
      the picture that Bill sold him
      the picture that Bill gave you
      the picture that Bill told me about
      the picture that Bill lent my wife
      the picture that Bill spilled Mazola all over
         that night when we all got stinking drunk
etc.

>In other words, 'user-interface' means a certain type of interface (the
>one for the user) as opposed to 'user interface' which stands for an
>aspect of the user, namely its interface.

Let's get away from the semantics here for a minute. Many hyphens
are needed to follow a rule of English syntax, which says that

   Noun modifiers [of more than one word] *follow* the noun they modify,
              whereas
   [One-word] Noun Modifiers *precede* it.

Otherwise known as the "eleven-year-old boy" rule, referring to the
minimal pair
             a boy eleven years old
             an eleven-year-old boy
where the second modifier is morphologically a single word, and its
hyphenation reflects that.  There's more to the rule, since English
can't have number marked on preceding adjectives; e.g, few visit a
shoe store to buy a single shoe, but *shoes store is impossible, ditto
*eleven-years-old boy.  So that's where a lot of the hyphens come
from.

If a compound noun phrase like "user interface" (which unambiguously
means "the interface [[to be] provided] for the user", I'm afraid) is
being used to modify another noun, as in the phrase "user-interface
design", then it'll be hyphenated to show that it has the following
parsing:  [[user interface] design], not [user [interface design]],
which could be spelled "user interface-design" and would mean some
design of interfaces by some user(s).  That almost works, except that
"user interface design" can be considered to be a three-word compound
noun and is equivalent to the first (but not the second) hyphenated
sense above.  Simple, right?

The real problem is that all noun compounding is idiomatic, in English
or any language.  There is a short list of common semantic relations
expressed, but human creativity being what it is, anything at all is
possible. And then we just have to cope with how we might want to
spell and punctuate it, given the limited resources of writing.

>Is this correct
>Or is it just the other way around?
>Or is it nothing like that?
>Can one use both forms?
>Or only if one of them does not make sense, so that
>it is obvious what is meant?

All of the above, for different people and situations.

Sorry I can't be more definitive.  Of course, you can always adopt
your own personal or company standard and see how far you can get with
it. Arbitrariness is a game that anybody can play.

  --------------------------------------------
 -John Lawler                     More grammar
  Linguistics Program   University of Michigan
 "..and, who knows? Maybe the horse will sing."