Relations that are seldom or never signaled

[fiorella de rosis]

>One of the aspects of RST that has received relatively little commentary or
>explanation concerns unsignaled relations in text.

A few years ago, we collected and examined a corpus of explanations about
drug prescriptions, from  a group of doctors in the UK. These explanations
were addressed to nurses, to patients and to colleagues (our purpose was to
get ideas on how to design a user-adapted explanation generation system).

We examined these texts in cooperation with a linguist in Reading, UK
(Richard Ingham: I don't know if he is in the rstlist). One particularly
interesting issue concerned the 'causal links' (a broad category including
Justification, Solutionhood, Purpose, which was very frequent in our
explanation texts): when these links could be inferred from the text, they
were not rendered with a 'so', a 'because' or another specific marker, but
with an 'and' or with no connective at all.

Some examples:
"I've examined Mrs Smith and I think she's got angina with some congestive
cardiac failure and I've prescribed some Frusemide".
"I've asked Mr Smith to come into hospital, he's suffering from
tuberculosis of the chest".
"This drug is a drug we've been using in practice for quite a long time
now, we know it's a safe drug".
...and many more similar.

It turned out that most of the examples in which the connective was
rendered with an 'and' or was omitted were from tests addressed to nurses
or to colleagues, while the connectives were usually rendered with a more
explicit marker in the texts addressed to the patients.

We came to the conclusion that the nature of the addressee should be seen
as a crucial variable in establishing whether and how to translate a RR
into a linguistic marker. Explanation texts usually include a main
proposition, and a background statement of the reason why this proposition
is true: if the speaker judges that the background information is well
known to he hearer, he will sometimes give only the proposition which, when
combined with what the hearer already knows, provides the explanation. Even
when the reason is given, the addressee is presumed to be able to make the
inference and the RR-related marker is omitted.

We claim that texts cannot be analysed (and generated) only according to
linguistic criteria, but that sociolinguistic and psychological aspects are
important to consider, as well: but, maybe, this viewpoint is linked to our
experience of non-linguists working in user-adapted interaction.