greatest interest to develop objective means of scoring
dreams that are independent of a human judgment and that
can be reproduced across laboratories. So far, automatic
analysis has not been used in studies of emotions in
dreams. The development of this technology could
improve our knowledge on dreams and be a major
breakthrough in this research area.
Sentiment Analysis in AI
In this work, we classify whole texts using 4-level scales.
In most related literature, texts are analyzed at the
sentence-level. This representation would be an interesting
alternative for our work but, unfortunately, the UofO
dream bank is not annotated at the sentence-level at this
time. Moreover, many works (e.g., Turney [12]) formulate
the problem as classifying texts as positive or negative
(binary classification). This formulation differs from our 4-
level scale that we motivate by the need of fine grain
analysis of sentiment strength for further processing (e.g.,
analyzing stress level of dreamers). We believe our
problem formulation is more difficult than the binary
classification but gives more flexibility.
The most severe limitation of our work is the rather limited
use of context. In [14], negations and modalities handling
is added to a model making use of the GI and the HM
lexicons. It allows recognizing when the context changes
the polarity of a word (for instance the passage “is not
kind” means the opposite of benevolent, charitable.) This
improvement is reported to future work items.
Conclusion and Future Work
In this paper, we show how to automate dream sentiment
analysis. We specifically experimented with techniques
aiming at rating a dream on a 4-level negative scale. We
reached accuracy of 50% with a mean squared error of
0.577, a statistically significant improvement over the
majority class guessing. We found that the GI and LIWC
resources offer the best features from an automatic dream
sentiment analysis point of view.
In our future work, we will first extend our dataset. We
expect that this will significantly improve our results,
given that we have a 4-class problem and only a very
limited set of labeled instances. We will also improve the
handling of negations and modalities that can completely
change the polarity of words in our current framework. The
long-term research goal would be to support further
processing in the dream analysis field such as stress
analysis.
Acknowledgement
We would like to thank Sonia Matwin who introduced us
to the LIWC resource.
References
[1] Delorme, M.-A., Lortie-Lussier, M., and De Koninck,
J. (2002). Stress and Coping in the Waking and
Dreaming States during an Examination Period.
Dreaming, 12(4), pp. 171-183.
[2] Domhoff, G. W. (2003). The Scientific Study of
Dreams: Neural Networks, Cognitive Development, and
Content Analysis. American Psychological Association,
New York.
[3] Hall, C. S. and Van de Castle, R. L. (1966). The
Content Analysis of Dreams. Meredith Publishing
Company, New York.
[4] Hartmann, E. (1998). Dreams and Nightmares: The
New Theory on the Origin and Meaning of Dreams.
Plenum, New York,
[5] Hatzivassiloglou, V. and McKeown, K. R. (1997)
Predicting the Semantic Orientation of Adjectives. Proc.
Conference of the European Chapter of the ACL.
[6] Hobson, J. A., Stickgold, R. and Pace-Schott, E. F.
(1998). The Neuropsychology of REM Sleep Dreaming.
NeuroReport, Vol. 9, pp. R1-R14.
[7] Maquet, P., Peters, J. M., Aerts, J., Delfiore, G.,
Degueldre, C, Luxen, A. and Franck, G. (1996).
Functional Neuroanatomy of Human Rapid-Eye-
Movement Sleep and Dreaming, Nature, Vol. 383,
pp.163-166
[8] Nielsen, T. A. and Strenstrom, P. (2005). What are
the Memory Sources of Dreaming? Nature, Vol. 437,
pp.1286-1289.
[9] Pennebaker, J. W., Francis, M. E. and Booth, R.J.
(2001) Linguistic Inquiry and Word Count LIWC2001.
Erlbaum Publishers, Mahwah, NJ.
[10] Stone, P. J., Dunphy, D. C., Smith, M. S., Ogilvie, D.
M. (1966) The General Inquirer: A Computer Approach
to Content Analysis, The MIT Press.
[11] St-Onge, M., Lortie-Lussier, M., Mercier, P., Grenier,
J. and De Koninck, J. (2005). Emotions in the Diary and
REM Dreams of Young and Late Adulthood Women and
Their Relation to Life Satisfaction. Dreaming, Vol. 15,
pp.116-128.
[12] Turney, P. (2002) Thumbs Up or Thumbs Down?
Semantic Orientation Applied to Unsupervised
Classification of Reviews. Proc. 40th Annual Meeting of
the Association for Computational Linguistics.
[13] Turney, P. and Littman, M. (2003) Measuring Praise
and Criticism: Inference of Semantic Orientation from
Association. ACM Transactions on Information Systems
(TOIS). 21(4). pp. 315-346.
[14] Wilson, T., Wiebe, J. and Hoffman, P. (2005).
Recognizing Contextual Polarity in Phrase-Level
Sentiment Analysis. Proceedings of HLT/EMNLP.
[15] Witten, I. H. and Frank, E. (2005) Data Mining:
Practical machine learning tools and techniques, 2nd
Edition, Morgan Kaufmann, San Francisco.