Tagging Psychotherapeutic Interviews for Linguistic 
Analysis 
Jon David PATRICK 
Basser Department of Computer Science 
University of Sydney, 
Sydney, NSW 2006, Australia 
jonpat@cs.usyd.edu.au 
Abstract 
A proposal is presented to create a tagsetfor 
linguistic features that constitute the meaning- 
ful elements to one particular theory of psy- 
chotherapeutic intervention, that is Grinder & 
Bandler' s Metamodel. These features include 
lexical, lexicogrammatical and semantic phe- 
nomena. It is proposed that determining the 
effectiveness of therapy could in some degree 
be automated by building an information sys- 
tem to automatically tag client interviews. An 
appraisal of the effectiveness of the computa- 
tional processing, as well as the therapy, might 
be established by comparing some characteri- 
sation of client interviews before and after 
therapy based on the features of the Meta- 
model. Three 30 minute interviews for each of 
10 clients are available, that is before and 
after group therapy sessions and 3-6 months 
later The prototype tagset is presented for 
appraisal. A hand analysis of some of the data 
gives prima facie justification for persuing this 
research. 
The appraisal of the effectiveness of psycho- 
therapy has proven to be extremely difficult. 
Both process and outcome studies have pro- 
vided incomplete evidence as to the effective- 
ness of therapy (Lambert & Hill, 1994). A 
recent review of the field states that not one out 
of hundreds of methods has been shown to be 
more effective than any other method, this 
includes the failure of psychotropic methods to 
be any better than "talk" therapies (Miller, 
Hubble, Duncan, 1995). This does not mean 
that therapy is not effective as it has been con- 
vincingly shown to be more effective than pla- 
cebo and control groups. However the 
effectiveness of therapy has been asserted to 
be most highly correlated with four contribu- 
tory factors; psyhcotherapeutic method 15%, 
therapeutic relationship 30%, client situation 
40%, client expectancy 15%(Miller, Hubble, 
Duncan, 1995). With this backdrop it can be 
seen that any method that can contribute to 
understanding and assessing the nature of 
change in a client will benefit not only the psy- 
chotherapy profession as a whole, but other 
professions with outcomes that require learn- 
ing by their clients. 
There are a great number of psychotherapy 
methods that dominantly require extensive 
conversational intercourse between therapist 
and client. Perhaps the most popularly known 
are psychodynamic, Cognitive Behavioural 
Therapy(CBT), and interpersonal therapies 
such as family therapy, narrative therapy, emo- 
tion-focussed therapy, solution-focussed ther- 
apy. Many of these therapies have developed a 
demand for linguistic analysis but with their 
strongest attention on the therapist's rhetoric 
and discourses. Their claims about client lan- 
guage fall more into the level of genre and ide- 
ology, that is the focus of the therapy is on 
varied aspects of the content not the lexico- 
grammatical structure of the language. The 
only method that has focussed dominantly on 
language structure has become known as Neu- 
rolinguistic Programming. It was originally 
proffered not as a therapy method but as a 
method of language analysis, called the Meta- 
model, suitable for use by practitioners of any 
therapeutic method. However with enhance- 
ments beyond its linguistic principles, it soon 
58 
became promoted as a method of therapy by its 
supporters. In that light Neurolinguistic Pro- 
gramming has to be viewed as different to 
other conversational therapies. It is not aligned 
with social constructivist models of realities 
that represent therapy as challenging and 
reworking socially constructed phenomena, 
and which are therefore computationally diffi- 
cult to evaluate. Rather it asserts an Intrapsy- 
chic focus where change is in the mind (and 
body) of the client and so has more potential 
for computational processing. 
The Metamodel method argues that humans 
have three principal mechanisms that are nec- 
essary for filtering information from the exter- 
nal world. They are generalization, deletion 
and distortion, from which people construct 
their model of the world. In communicating 
with each other, humans project their model of 
the world, hence any impoverishment in this 
model will be produced in their language. In 
the circumstances where people are suffering 
emotional pain brought about by limitations in 
their model, and therefore their sense of not 
having choices of action, a search of their lan- 
guage structures should show evidence of gen- 
eralization, distortion and deletion, indicating 
the impoverished components of the model. 
The process of psychotherapeutic change then 
begins with detecting the impoverishments and 
then assisting the client to recognise and then 
recover the missing pieces. This places the cli- 
ent in the position of active self-discovery, 
expanding themselves by actively expanding 
their model of the world (Bandler & Grinder, 
1975). 
The Metamodel method is claimed to have the 
characteristics of, firstly, specifying a process 
for moving from the surface structure of an 
utterance to its full semantic content, secondly, 
it provides the therapist with an accurate view 
of the client's model of the relevant part of the 
world, and thirdly, it provides a format for 
challenging the full semantic structure of an 
utterance and reconnecting it with direct per- 
sonal experience. These characteristics of the 
Metamodel mean that the therapist can chal- 
lenge the client's assumptions that their lin- 
guistic model is reality, which is a major part 
of the therapeutic process. 
The linguistic classes that are explicitly listed 
for use by the Metamodel are the following: 
Generalization - generalized referential index, 
non-specific nouns, non-specific verbs, univer- 
sal quantifiers, 
Distortion - reversed referential index, nomi- 
nalisations, some adverbial constructions 
(obviously, clearly, surprisingly) 
Deletion - deleted referential index, non-spe- 
cific referent, comparatives and superlatives 
with missing referents, modal operators of 
possibility and necessity, missing information, 
presuppositions, semiotic distance (use of pas- 
sive voice) 
Semantic ill-formedness - cause & effect, 
implied causatives, mind reading, lost perfor- 
matives and complex equivalences. 
Other linguistic classes were added with the 
development of the idea of Representational 
Systems, that is, a biased use of sensory spe- 
cific words of the visual, auditory, kinaesthetic, 
olfactory and gustatory classes (Grinder & 
Bandler, 1976) 
These classes are lexical, lexicogrammatical 
and semantic. We are interested in processing 
text to identify and tag each of these phenom- 
ena. A set of tagged examples covering these 
linguistic classes is presented in Appendix 1 
and a prototype tagset is presented in Appen- 
dix 2. The proposed tagset at this stage 
attempts to capture the requirements of the 
Metamodel at a macro level. It is accepted that 
this set will be subject to change and refine- 
ment as a better understanding of both the lin- 
guistic and computational problems emerge, 
and in particular the linguistic theory is 
updated to include more contemporary knowl- 
edge. 
59 
The Metamodel method has the concept of 
statements that are "well-formed in therapy" 
with seven criteria for assessment, which 
might also be used as criteria for characteriz- 
ing client dialogues. They are: 
- well-formed in english 
- contain no transformational deletions or 
unexplored deletions of areas of the model 
in which the client experiences no choices. 
- contains no nominalisations 
- contain no words or phrases lacking referen- 
tial indices 
- contain no verbs incompletely specified 
- contain no unexplored presuppositions in the 
portion of the model in which the client 
experiences no choice 
- contain no sentences which violate the 
semantic conditions of well-formedness. 
For the purposes of assessing the value of a 
psychotherapy program, it is conceivable that 
the change in linguistic usage by a client repre- 
sents the effectiveness of the psychotherapeu- 
tic process. If such a postulate is accepted then 
the analysis of the comparison of conversa- 
tional speech of the client before therapy and 
after therapy would be an appropriate measure 
of therapeutic effectiveness. 
In our case we have 10 clients with 30 minute 
interviews recorded before therapy, after ther- 
apy and 3-6 months downstream of therapy 
totalling 15 hours of transcripts. We intend to 
develop automated processing methods to 
identify the linguistic classes described by the 
Metamodel, construct a characterisation of 
each interview and test for a change in lan- 
guage usage by the clients. A collation of a few 
classes of representational system data per- 
formed by hand gave some encouragement that 
this approach is useful. It showed that 7 out of 
10 clients had a significant change in their lan- 
guage towards the desirable direction, that is 
greater diversity. Two showed no change and 
one showed retrograde change, that is nar- 
rower diversity. These results were consistent 
with the therapist's observations about the 
progress of the clients in their therapeutic pro- 
gram. 

References 
Bandler, R. & Grinder, J. (1975). The Structure 
of Magic. Palo Alto: Science and Behavior 
Books. 
Grinder, J. & Bandler, R., (1976). The Struc- 
ture of Magic H. Palo Alto: Science and 
Behavior Books 
Lambert, M. J., & Hill, C. E. (1994). Assessing 
psychotherapy outcomes and processes. In 
A.E. Bergin & S. L. Garfield (Eds.) Handbook 
of Psychotherapy and Behaviour Change, (4th 
ed.), pp 72-113. New York: John Wiley & 
Sons. 
Miller, S. Hubble, M. Duncan, B. (1995). "No 
More Bells and Whistles". Networker, March/ 
April, pp 53-63. 
