Bashar Nuseibeh, Paul Piwek, Anne De Roeck, Alistair Willis
Department of Computing
The Open University, U.K.
Acknowledgement: MaTREx Project (EPSRC Grant Number: EP/F069227/1)
Research Aim and summary
To discover tacit knowledge in requirements text by
tracking linguistic presuppositions.
“Nocuous tacit knowledge”
Determined by human judgement
MaTREx project on use of NLP techniques, such as ambiguity
analysis and presuppositions
“We know more than we can tell” – Polanyi, 1963
Janik (1988) argues that the term “tacit knowledge” is used at two ways:
firstly, following Polanyi, tacit knowledge is knowledge inexpressible in
words, and it is acquired by familiarity or practice such as smells and
secondly, tacit knowledge at a shallow level is knowledge not yet put into
words such as craft knowledge and presuppositions.
We adopt Janek’s second perspective:
Tacit knowledge is knowledge that knowers know and could have
articulated but omit doing so for some reason, perhaps because they
simply were not asked.
Presuppositions are background information or
assumptions that can be taken for granted.
The King of France is Bald.
Presupposition: There is king of France.
John knows that Susan is coming to the party.
Presupposition : Susan is coming to the party.
Richard managed to pass the exam.
Presupposition : Richard tried to pass the exam.
Presupposition is believed to be signalled by certain
types of syntactical structure, which are called
1. Definite description: The King of France
2. Factive verb: know
3. Implicative verb: manage
Presupposition triggers - cont.
The trigger types include:
Definite descriptions, e.g. the device, its accessibility;
Factive verbs, e.g. know, reveal;
Implicative verbs, e.g. avoid, intend;
Change of state verbs, e.g. continue, stop;
Clefts – it + be + noun + subordinate clause;
Stressed constituents – words in italic in texts;
Counter factual conditionals – what would be the case is
something were true;
Expressions of repetition, e.g. also, too;
Temporal relations, e.g. since, after;
Comparisons, e.g. less/larger than
Our preliminary case study
We studied a 20-page
requirements document for
integrated circuit chip design.
Our study was mostly
manual, although we
automated the identification
of noun phrases.
We recorded the kinds of
appeared, and found the
majority triggered by definite
Examples found in document
Sentence: “Accessibility in the experimental hall is required for changing the
piggy board where the device will be mounted.”
Presuppositions: There is a piggy board.
There is a device.
Sentence: “…tests revealed that redundancy to Single Event Upsets is
Presupposition: Redundancy to Single Event Upsets is required.
Sentence: “…chambers shall avoid that two CMA share the same gas
Presupposition: Two CMA may share the same gas volume.
Which presuppositions are “dangerous”?
“Accessibility in the experimental hall is required for
changing the piggy board where the device will be
A new device or “the piggy board”?
“...will have various interfaces for different groups of
users. While the appearance of the user interface may
be similar, the functionality of each user interface will be
“The user interface” refers to “various interfaces” or
“each user interface”?
Nocuously Tacit Knowledge
“danger” is in the eye of the beholder (the reader).
One way to determine this is by conducting empirical
studies to elicit human judgements (ala Chantree et al
As with nocuous ambiguity, nocuous presuppositions
are those that signal tacit knowledge who tacitness may
have a negative impact on the reading interpretation of
What we know:
By using natural language processing techniques, definite
descriptions can easily be found.
Where we are:
Currently, there are only a few representative example words or
constructions of presupposition triggers. They can only be found
What we need to do:
Detect more presupposition triggers based on natural language
processing techniques, and try to relate these to significant tacit
Automatically tracking presupposition by NLP
K. Wiemer-Hastings and P. Wiemer-Hastings, “DP: a detector for
presuppositions in survey questions,” Proceedings of the sixth conference
on Applied natural language processing, 2000, pp. 90–96.D.
Clausen and C.D. Manning, “Presupposed Content and Entailments in
Natural Language Inference,” ACL-IJCNLP 2009, p. 70.
Nothing in RE?
Future Work (Lin’s PhD research agenda!)
Case study on behaviour and linguistic attributes of
presuppositions in more requirements documents with
the help of NLP.
Discovery of nocuous presuppositions by collecting
human judgments from stakeholders
Building a system to automatically highlight
presuppositions that have negative impact on
communication in requirement documents.
Our preliminary work has shown that tacit knowledge
can be extracted by tracking presuppositions in
With the help of NLP techniques and the involvement of
human judgements, tracking presuppositions in
requirements can make some elements of tacit