A Phase Transition of the Unconscious: Automated Text Analysis of Dreams in Psychoanalytic Psychotherapy

Abstract

Aim: Psychotherapy could be interpreted as a self-organizing process which reveals discontinuous pattern transitions (so-called phase transitions). Whereas this was shown in the conscious process of awake patients by different measures and at different time scales, dreams came very seldom into the focus of investigation. The present work tests the hypothesis that, by dreaming, the patient gets progressively more access to affective-laden (i.e., emotionally charged) unconscious dimensions. Furthermore, the study investigates if, over the course of psychotherapy, a discontinuous phase transition occurs in the patient's capacity to get in contact with those unconscious dimensions.

Methods and procedures: A series of 95 dream narratives reported during a psychoanalytic psychotherapy of a female patient (published as the "dreams of Amalie X") was used for analysis. An automated text analysis procedure based on multiple correspondence analysis was applied to the textual corpus of the dreams, highlighting a 10-factor structure. The factors, interpreted as affective-laden unconscious meaning dimensions, were adopted to define a 10-dimensional phase space, in which the ability of a dream to be associated with one or more local factors representing complex affective-laden meanings is measured by the Euclidean distance (ED) from the origin of this hyperspace. The obtained ED time series has been fitted by an autoregressive integrated moving average (ARIMA) model and by non linear methods like dynamic complexity, recurrence plot, and time frequency distribution. Change point analysis was applied to these non linear methods.

Results: The results show an increased frequency and intensity of dreams to get access to affective-laden meanings. Non linear methods identified a phase transition-like jump of the ED dynamics onto a higher complexity level of the dreaming process, suggesting a non linear process in the patient's capacity to get in contact with unconscious dimensions.

Conclusion: The study corroborates the hypothesis that, by dreaming, the patient gets progressively more access to affective-laden meaning intended as unconscious dimensions. The trajectory of this process has been reproduced by an ARIMA model, and beyond this, non linear methods of time series analysis allowed the identification of a phase transition in the unconscious process of the psychoanalytic therapy under investigation.

Keywords: dream analysis; meaning; phase transition; psychotherapy process; text analysis.

FIGURE 1

(A) Binary coded time series of dreams within the centroid (value 1) or outside the centroid (value 2) of the 10-dimensional factor space. (B) Dynamic complexity of the binary coded time series (A) (window width: 7).

FIGURE 2

(A) Autocorrelation function (ACF) and (B) partial autocorrelation function (PACF) graphs. The straight lines above and below the 0.0 line indicate the 95% confidence interval (CI). ACF and PACF values (black bars) which clearly exceed the CI line appear only at lag 1, suggesting an AR1 (p = 1) and MA1 (q = 1) model.

FIGURE 3

Time series of the Euclidean distances (EDs) of each dream from the origin in the factor phase space (black line) and the curve of the fitted ARIMA_(1,1,1) model (black dotted line) within a confidence band (dotted gray lines: upper and lower limits of the 99% confidence interval).

FIGURE 4

(A) Autocorrelation function (ACF) and (B) partial autocorrelation function (PACF) of the residuals of the ARIMA_(1,1,1) model. The straight lines above and below the 0.0 line indicate the 95% confidence interval (CI). The correlation values (gray bars) lie within the 95% CI limits, which indicates that the errors of the residuals are white noise. This proves that the model is appropriate for prediction.

FIGURE 5

(A) Time series of the Euclidean distances of each dream (compare with the black curve in Figure 3). The straight vertical line indicates the average of the change points identified on the ED time series by CPA. Criterion of the CPA: changing variance. (B) CPA (blue dot) applied to the ED time series. (C) CPA (blue dot) applied to the DC (window width: 7) of the ED time series. (D) CPA (black vertical line) applied to the RP of the ED time series. Parameters: Three embedding dimensions and τ = 1. CPA was applied to each line and column of the RP. (E) CPA (black vertical line) applied to the TFD of the ED time series. The red colors indicate the highest amplitudes of the time-dependent frequency distribution. The average of the change points is at dream 58 [black line in panel (A)].

FIGURE A1

Representation of a sample of dreams related to the mean of the EDs (centroid). The x- and y-axis represent factor 1 and factor 2, respectively. The black line represents the centroid.