Relations that are seldom or never signale

[Bill Mann]

Responding to Manfred and Claudia about unsignaled Concession:

In one of the early papers there was a Concession example as follows:

You have excellent credentials.

I am looking for someone with excellent experience.

This example came up in the midst of a test of whether absolutely every relation
that we had defined could appear unsignaled.  We finally said yes, every one
could appear unsignaled.  (Unfortunately, to complete the test, we had to
include a small number os constructed examples.  This was one of them.)

Still, I think that every one can appear unsignaled.

Bill Mann



______________________________ Reply Separator _________________________________
Subject: Re: [RST-LIST] Relations that are seldom or never signaled
Author:  <soria at ILC.PI.CNR.IT> at Internet
Date:    1/14/00 6:45 PM


Hi RSTlisters,

Manfred Stede wrote:

> The discussion on relations seldom/never signalled reminds me of the
> reverse question: Are there relations that are always signalled?

According to an experiment of mine, the "Antithesis", "Concession",
"Otherwise" and "Contrast" relations are almost always signalled (> 90%).
(In this experiment, I compared the rhetorical structure of two sets of
narratives produced by ten subjects. Each subject had to produce a spoken
and a written version of the same story, which had been presented under
the form of a set of drawings.)
The interesting result was that different types of relations were marked
with a different frequency:  in general, relations such as Elaboration,
Restatement or Interpretation tend to be marked less than relations such
as n-v Cause, n-v Result, Purpose, Evidence, Justify or Solutionhood,
while relations such as Concession, Contrast etc. are marked in almost all
cases.

In addition, there seemed to be a difference between speech and writing
for what concerns the frequency of marking: spoken language tends to use
more markers than written language does, although the difference is not
significant (S: 48,4%, W: 37,3%)

However, as Maite points out, it's clear that people rely on many cues
other than connectives to infer coherence relations; actually, most
instances in my data were marked by no connective at all (51,6% for spoken
language; 62,7% for written language).


> A
> personal hobby of mine is CONCESSION, for which I'd be particularly
> curious to hear about unsignalled examples.
> (My guess being that the cognitive load of the CONCESSION operation,
> which I think involves overwriting default assumptions, is too high to
> be triggered without an explicit instruction. Hence, an "empty" marker
> such as 'and' doesn't suffice - yet the relatively unspecific and
> "somehow contrastive" 'but' can signal the CONCESSION.) Do people have
> thought on this?

I do. I completely agree with this point. In the above mentioned
experiment, it resulted that all relations involving a negative component
(such as the Concession relation) are always marked by a connective. My
guess was exactly the same: in order to perform the cognitive operation
needed by a Concession relation, namely to cancel a default implication,
an explicit marker is in order.
However, since in my data there were instances of contrastive relations
which were not marked, it would be interesting to verify Manfred's
assumptions that i) a connective is always needed and ii) this connective
cannot be of the "weak" type such as "and".

Cheers,

Claudia