Text Mining Descriptions Of Dreams: aesthetic and clinical efforts

10/26/2017
by   Renato Fabbri, et al.
0

Dreams are highly valued in both Freudian psychoanalysis and less conservative clinical traditions. Text mining enables the extraction of meaning from writings in powerful and unexpected ways. In this work, we report methods, uses and results obtained by mining descriptions of dreams. The texts were collected as part of a course in Schizoanalysis (Clinical Psychology) from dozens of participants. They were subsequently mined using various techniques for the achievement of poems and summaries, which were then used in clinical sessions by means of music and declamation. The results were found aesthetically appealing and effective to engage the audience. The expansion of the corpus, mining methods and strategies for using the derivatives for art and therapy are considered for future work.

READ FULL TEXT VIEW PDF

page 1

page 2

page 3

page 4

07/27/2022

The Leaf Clinical Trials Corpus: a new resource for query generation from clinical trial eligibility criteria

Identifying cohorts of patients based on eligibility criteria such as me...
04/10/2020

Automated Spelling Correction for Clinical Text Mining in Russian

The main goal of this paper is to develop a spell checker module for cli...
04/30/2018

Q-Map: clinical concept mining with phrase sense disambiguation

Over the past decade, there has been a steep rise in data driven analysi...
06/30/2016

Relation extraction from clinical texts using domain invariant convolutional neural network

In recent years extracting relevant information from biomedical and clin...
07/09/2020

DISCO PAL: Diachronic Spanish Sonnet Corpus with Psychological and Affective Labels

Nowadays, there are many applications of text mining over corpus from di...
12/10/2020

A Practical Approach towards Causality Mining in Clinical Text using Active Transfer Learning

Objective: Causality mining is an active research area, which requires t...
10/26/2021

Diachronic Text Mining Investigation of Therapeutic Candidates for COVID-19

Diachronic text mining has frequently been applied to long-term linguist...

1 Introduction

Although dreams are described in texts that range from ancient sacred (Barker et al., 1993; Boas & Boas, 1974; Kopenawa, 2013) to psychiatric (Esposito et al., 1999), there is no consensus of what dreams are. We can exemplify the diverse theories with three simple cases (Domhoff, 2013):

  • dreams are often regarded by the dreamers as accessing spiritual realms and other realities.

  • Many scientists regard dreams as by-products of the sleeping process: arbitrary interpretations given by the conscious mind to noisy signals without substantial meaning.

  • Freudian and Jungian psychoanalytic traditions understand dreams as symbolic constructs output by the unconscious mind.

We can, however, state some facts about dreams that assert them as attractive for therapy and for art. First, dreams are often very rich in impacting and symbolic images. Second, they are told by the person who dreamed in a very attentive manner, as being very significant to the dreamer. In fact, most of us should be able to recall a number of situations where someone (perhaps ourselves) was describing a dream in a rapid, almost euphoric, succession of words. Dreams are so effective in yielding artistic materials that surrealism is an aesthetic explicitly inspired by dreams and symbolism is an example of artistic movement heavily influenced by dreams.

Text mining is data mining applied to textual data. There are many models for the text mining pipeline, but it can be summarized as: data collection and preparation, pattern recognition, evaluation of the output and reporting 

(Fabbri, 2017). This work addresses text mining of descriptions of dreams with aesthetic and therapeutic purposes.

Section 2 describes the corpus and methods. Section 3 is dedicated to presentation and discussion of results. Section 4 holds conclusions and further work considerations.

2 Materials and Methods

2.1 Corpus

The description of dreams we used are all in Brazilian Portuguese, collected as part of clinical practices in the year of 2015. Participants described the dreams and sent them to the second author of this paper, who is a psychologist. Thereafter, another collection of dreams was gathered in the same way by the second author and collaborators, with the purpose of expanding the analysis and synthesis of texts performed with the previous corpus. It is a larger corpus, also in Brazilian Portuguese. Interestingly, both corpus contain descriptions of dreams by women only. Their scales are summarized in Table 1 in terms of numbers of characters, tokens, and paragraphs.

File Characters Tokens Sentences Paragraphs Date
corpora.txt 9456 1693 104 30 Mar/2015
corpora2.txt 71514 14691 435 156 Nov/2015
Table 1: Corpus files and elementary countings. The number of dreams is about the same as the number of paragraphs. The date column corresponds to the month when the last dream was collected.

2.2 Analysis and derivation methods

The texts were analyzed to support the extraction of meaning from the dreams and for the creation of artistic texts. We strived to keep the methods very simple in order to avoid puzzling the involved parties. Three lists of tokens were considered:

  • punctuations !”#$%&’()*+,-./:;¡=¿?@[]\^_‘{}—~. Obtained through the command
    string.punctuation of Python’s string (standard) library.

  • Portuguese stopwords111The exact definition and list of stopwords are not consensual. Anyway, one can regard them as words with lesser meaning and which are very frequent, such as conjunctions and prepositions. obtained through NLTK (Bird, 2006) by the command
    nltk.corpus.stopwords.words("portuguese").

  • Tokens in the texts which were not punctuations nor stopwords. These were regarded as the most meaningful words in the corpus.

This selection of most meaningful words was used as the core material for the achievement of more interesting constructions for art and clinical psychology through filtering and ordering. Most significantly, the ordering could be based on the alphabet, the size of tokens in number of letters, or the count of incidences of the words, or any combination of these. Filtering could be performed by restricting the vowels, consonants (e.g. fricatives), word size, frequency, or collocations.

3 Results and Discussion

The list of most meaningful words (described in the previous section) was filtered and ordered in many ways to yield diverse sequences of interest. After an inspection of the results, these criteria were selected to compose a final document:

  • Ordering by: incidence (most frequent words first), alphabetic, size in characters, with and without repetitions. These were considered the most raw sequences and used subsequently to derive other sequences with such variations of ordering and repetition.

  • Words with only one vowel (repeated any number of times).

  • Only words with fricatives or plosives or some combination of them (e.g. plosives and ’m’ and vowels ’a’ and ’e’).

  • Words that start and end sentences.

  • Collocations (pairs of words which are frequent together).

Such final document and other files are available online and exposed in Table 2. An example of the derived texts is in Table 3 with a translation from Portuguese to English. These texts were used for aesthetic appreciation and also in a schizoanalysis group session in 2014 (Casa Nuvem, Rio de Janeiro, RJ, Brazil). In the same course, the artist Giuliano Obici used the texts to feed his program ”Voices Simulacrum” (Obici, 2015): a machinic chorus of robotic voices with computers connected in network. The project integrated devices in the computer (sound, video and network cards) exploring the network as a distributed audiovisual instrument (conceptualized as a “metamedia-instrument”) to read the texts in a very performative manner.

The group was constituted by the participants who described the dreams and the report of the episode is somewhat impressive: the members had strong impressions, some of them cried and entered a quasi-shock state.

File Description
scripts/todos.py Python script that makes the current analysis and renders the TXT and PDF files.
corpora/corpora.txt The first collection of descriptions of dreams.
corpora/corpora2.txt The second (and larger) collection of descriptions of dreams.
mineracaoDosSonhos/PLNSonhos.odt A brief consideration of the text mining of dreams to which this article is dedicated.
mineracaoDosSonhos/TUDO.pdf A thorough exposition of all the (selected) texts derived from the descriptions of dreams.
Table 2: Files related to the text mining of dreams. All files are found in a public git repository dedicated to the developments presented in this article (Fabbri & Borges, 2015).
Portuguese (original) English (translation)
Escorregava glandes Slipping glands
Numa assustavam At once, they scared
Eu suada I sweated
As cavalos The horses
Não acabou It’s not over
Barras mim Bars me
Andei construtores I walked builders
Pessoas ) People )
Sonhei formei I dreamed I formed
Estava menino It was boy
Depois boa Then good
Esse meu This mine
Sonhos descendo Dreams coming down
O irmão The brother
Meu punição My punishment
Começa irmão Begins brother
Meu ele My him
Meu demonstração My demonstration
Depois ” After ”
Eu parede I wall
Sinto dele I’m fell him
A importência The “importence”
O buraco The hole
Acordei ofegante Woke up breathless
Sensação NÃO Feeling NO
Já rumo I’m on my way
Estava perseguido Was persecuted
Quando percebeu When realized
A tempo In time
Até porta Up to door
O disso The this
Eu sobreviver I survive
Parecia ferramentas Seemed like tools
Três mim Three of me
Lavo piano I wash piano
Havia tirano There was a tyrant
Nele tudo In him all
Jogaram fogo They set fire
Pessoas presas People trapped
Comecei ali I started there
Mas destruísse But destroy
Pessoas criança Children people
Table 3: Example of artistic text achieved from the descriptions of dreams. This text was obtained through picking only the first and last words of each sentence. As illustrated in this text, the unusual (and formally wrong) morphological and syntactic constructions were used for enhanced artistic expression and as clinical evidence of complex cognitive elaborations.

4 Conclusions and Future Work

We understand that the results are compelling for both art and clinical psychology. Only the first corpus was used, which is smaller and made easier the selection of the resulting texts. The methods applied are very simple, favoring the communication between the parties, and are promptly deepened and expanded into more complex processes. This work seems unique in the sense of using text mining of dreams for art and clinical psychology, which, in our opinion, benefits the appreciation of it as a multidisciplinary and scientific contribution in computer science, art and psychology.

In further efforts, we might use for the corpus:

  • descriptions of dreams in the literature (e.g. from the mythology, traditional communities, etc);

  • other languages;

  • an expansion of current corpus;

  • dreams from specific groups, e.g. again gender related or of a specific age span, professional or educational background, etc.

About the text mining methods, we might:

  • use specific routines for classification (e.g. clusterization) of texts or their features;

  • expand the methods of selection of words to better encompass meter (e.g. number of syllables);

  • expand the methods of selection of words and phrases by their sonorities (e.g. by using sequences of vowels or consonants, mute consonants, paroxytones);

  • use Wordnet (Fellbaum, 2010) in order to relate terms through semantic links (e.g. hypernymy, meronymy, synonymy).

The exploration of the results in therapeutic sessions and for the achievement of collections of artistic texts should be kept as the core purposes.

Acknowledgements

The authors thank the volunteers who supplied the descriptions of dreams; the subjects who attended to the schizoanalysis sessions; the open source software developers, especially those who enabled this work by developing the Python language and the NLTK.

References

  • Barker et al. (1993) Barker, K. L., Burdick, D., Burdick, D. W. (1993). The NIV study bible, new international version. Zondervan Bible Publishers.
  • Bird (2006) Bird, S. (2006). NLTK: the natural language toolkit. In Proceedings of the COLING/ACL on Interactive presentation sessions (pp. 69-72). Association for Computational Linguistics.
  • Boas & Boas (1974) Boas, O. V., & Boas, C. V. (1974). Xingu: the Indians, their myths. Souvenir Press.
  • Domhoff (2013) Domhoff, G. W. (2013). Finding meaning in dreams: A quantitative approach. Springer Science & Business Media.
  • Esposito et al. (1999) Esposito, K., Benitez, A., Barza, L., Mellman, T. (1999). Evaluation of dream content in combat-related PTSD. Journal of traumatic stress, 12(4), 681-687.
  • Fabbri (2017) Fabbri, R. (2017). Topological stability and textual differentiation in human interaction networks: statistical analysis, visualization and linked data. PhD thesis, University of São Paulo. https://github.com/ttm/thesis/raw/master/thesis-rfabbri.pdf
  • Fabbri & Borges (2015) Fabbri, R., Borges, F.M. (2015). Public git repository with the texts, scripts and corpora for the analysis of dreams. https://github.com/ttm/sonhos
  • Fellbaum (2010) Fellbaum, C. (2010). WordNet. In Theory and applications of ontology: computer applications (pp. 231-243). Springer Netherlands.
  • Kopenawa (2013) Kopenawa, D. (2013). The falling sky. Harvard University Press.
  • Obici (2015) Obici, G. (2015). Vozes Simulacrum: a machinic choir with connected computers. From http://www.giulianobici.com/site/vozessimulacrum.html