You are here

DHIL blog

News about the Digital Humanities Innovation Lab

Dreams, Narratives… Analysis

Dream Language Word Cloud
Published March 28, 2018 by Rebecca Dowson

Most people regard their nocturnal adventures a little higher than sheer gibberish, and they tend to quickly forget about them a few minutes after waking up. At Dr. Tore Nielsen’s Dream and Nightmare Laboratory in Montréal though, these are highly valued as items of a scientific inquiry into consciousness and cognition. Talk of text analysis, at first, might seem a bit out of place here; dreams are hallucinatory experiences that take place during sleep - they don’t seem to be the kind of things that give way to objective study. But precisely because of this, there is a long tradition in philosophy and psychology to treat dreams as one and the same thing with dream reports: having a dream is just telling a dream. And if dreams are narratives, then they can be subject to similar kind of inquiries that literary narratives are subject to. This is where text analysis comes into play.

Digital text analysis - a loose collection of methods for creating insight from variety of corpora using computer algorithms - is a key approach in digital humanities. Using the automation power of machines, narratives that amount to millions and millions of lines of text can be studied to reveal aspects that are practically impossible to catch by bare eye. Common lines of inquiry consist of sentiment analysis - extracting affective information from the corpus -, and topic modeling - discovering semantic relations hidden in the text -, among others.

During last two months I had the chance of visiting Dr. Nielsen’s lab and working together with Dr. Elizaveta Solomonova, an alumnus of the lab and currently a post-doc at McGill Pscychiatry. Dr. Solomonova’s work investigates the effect of environment and learning on dreaming. More precisely, she is looking for how external stimulation -such as a pressure cuff applied on a leg - during sleep is incorporated into the dream narrative, as well as neighbouring topics such as whether people with heightened bodily senses - such as Vipassana meditation practitioners - experience their dreams any differently. There are several dimensions one can look for to see such effects. One such dimension is to track the changes in the valence of the dream. For instance, the effect of introducing a foul-smelling stimulus during REM sleep can be observed in the subsequent dream report, as a narrative expression of disgust. This is a task that naturally lends way to sentiment analysis. Another is to look for mentions of semantically related textual items. If a pressure cuff is applied to a sleeper’s leg, one can expect to find a dream theme related to imprisonment or escape - an inquiry that can be carried out with topic modelling.

In such laboratory studies, there are - as of yet - no standardized metrics to evaluate results. Moreover, since evaluations are carried out by humans, studies are limited to smaller corpora. In order to approach objectivity though, standardized metrics as well as formalized tools are needed, and this requires analysis of larger corpora to discover fundamental patterns. This is a side goal of the project. To do so, we have decided to use digital text analysis in coming up with a formal method. We are working primarily with Python’s Natural Language Toolkit - a free collection of libraries that are used by people from various disciplines concerned with analyzing natural language. Python is an easy to learn programming language, but some of the techniques used in digital text analysis also have more accessible resources on the net, such as Google’s Ngram Viewer and Voyant Tools. The first stage of work begins with dictation of collected dream reports into digital format. Here we also eliminate the idiosyncratic language and normalize the grammar to make the corpus ready of NLTK library. After this, text clustering is done to divide the corpora into smaller semantic units. With clustering it becomes possible to ‘play’ with the data using various methods and algorithms that comes with NLTK. Our work is in progress, but to our knowledge this is the first such attempt in stimulus incorporation studies in dream science.

Fnd out more about The Dream and Nightmare Lab and Dr. Solomonova’s work.

Doğan

Blog Categories