What were we thinking? Is it possible to discover how past cultures made decisions, prioritised issues, or which ideas were felt to be emotional or bland, offensive or pleasant?
Language provides a fossil record of society, and big data has made huge progress in making historical psychology through language analysis possible and accessible. A step change occurred in 2010 when Google released frequency counts of over 500 billion words published in books from 1500 to the present day, sparking a new field of culturomics.
But frequency doesn’t tell us how languages change, just that they do. Figuring out how precisely words, and the ideas they represent, change requires painstaking examination of contexts in which words occur, or broad-brush studies of general properties of change.
One well-worn example is the meaning of the word “nice”. Google books tells us it increases in frequency from 1800 to 2000. Hand-coded examples in historical linguistics show that it changed from referring to foolish in the 1500s, to persnickety in the 1700s, before taking on its current meaning of, well, nice. We can then see that the increase in frequency actually corresponded to changes in use of “nice” as a positive rather than negative term (see the figure below).
The figure plots the change in valence (negative is low, positive emotion is high) and frequency of the word “nice” from 1800 to 2009.
In an article published in the special issue of Behavior Research Methods on Big data, Tomas Engelthaler, Cynthia Siew, and Thomas Hills have, for the first time, made this study of meaning change over time automatic, with a suite of tools that they call the Macroscope. The figure above was produced by macroscope.tech.
The Macroscope exploits a key characteristic of language, namely that we “know a word by the company it keeps,” as Firth stated. So, to figure out a word’s meaning from the past, we need to know the contexts in which it occurred in the past. Using the Google books corpus, co-occurrences between words can be automatically determined at each snapshot of time, and consequently changes in a word’s meanings can be plotted over time.
For the example of “nice”, the Macroscope fills in the background picture. In 1800, “nice” was closely related to abstruse, discussion, disquisition, and delicate. By 1990, the Macroscope shows connections to pretty, cute, and funny. Determining whether these closely-connected words are positive or negative confirms that “nice” does increase in positive sentiment over the last two hundred years. The figure below shows the drift of meaning of the word “nice” from 1850 (right upper quadrant) to 2000 (left upper quadrant) as produced by macroscope.tech.
Two more illustrations, also explored by Li and colleagues, show the power of the Macroscope to reflect historical psychology. In terms of frequency, the word “gay” declines in frequency into the 1980s but then increases rapidly in frequency. The Macroscope shows that gay moves from a meaning related to joy, colour, and flowers, to a meaning related to political movement, human rights, and sexuality by 2000. The historical psychological change corresponds with the gradual introduction of equal rights for homosexuals, and association with a broader rights movement.
The word “nuclear” changes from a context of physics terminology in the 1950s to being weaponized by 2000. These examples show that cultural change is reflected in the subtle, but tractable, adjustments in interconnections between words over time.
Of course, linking language meaning to cultural events cannot be directly inferred, but taking language as one symptom of societal change, and ensuring better tools to investigate those symptoms, means we can get closer to uncovering the history of our psychology. In an email discussion over his paper, Thomas Hills notes,
“Like any other area of psychology, the key challenges for historical psychology are data quality, validation, and the eventual weight of the evidence established using different approaches and methodologies.”
So, what is the future for historical psychology? In our email discussion, Thomas Hills suggests the next big topic is to further investigate what drives change in society and the information it produces as revealed through changes in the meaning of language. He asks,
“How does historical language track other kinds of historical indicators, such as subjective wellbeing, GDP, inequality, political ideologies, war, and so on?”
Aligning language with other cultural indicators over time will provide more detailed tests of claims, and generate the context that every archaeological discovery requires. Hills continues,
“Because the Macroscope contains many words and ideas, we can look at how those ideas change in aggregate in response to all kinds of things. What words are most likely to die, become more positive, or change their meanings over time? And how does their context influence that change?”
The Macroscope’s enriched representation of meanings over time mean that the flesh, and not just the bones, of language can be recreated.
And what about the future more generally? The Macroscope reveals that the “past” has been seen positively and this has been stable over time. But the “future” is much more volatile, with a peak in positivity around 1900, after which the world wars of the 20th century dealt a blow to optimism. Nevertheless, the future is on the rise again in positivity from 1990 onwards. At least, that is, until the current most recent date of the Macroscope data, in 2009.
Psychonomics article considered in this post:
Li, Y., Engelthaler, T., Siew, C. S., & Hills, T. T. (2019). The Macroscope: A tool for examining the historical structure of language. Behavior Research Methods, DOI: 10.3758/s13428-018-1177-6.