Words that convey strong emotions often involve bodily organs (e.g. my heart aches, I have butterflies in my stomach), and we speak of kind people as being warm. Language is full of embodied concepts like these.

Many of these concepts make sense in the real world. Concrete words like “attic” and “basement” orient our attention toward the spatial location in which they would normally occur. Thus, when doing a semantic relatedness tasks, presenting “attic” above “basement” helps participants determine that they are related, but having “basement” above “attic” hurts linguistic processing. That is, these words are said to participate in spatial and conceptual cuing because they have a prototypical spatial location.

At the same time, language allows us to express abstract ideas and to talk about things that we cannot see, such as “delays” and “dreams”. Surprisingly, we also associate many abstract words like these with directions such as “up” or “down”. Researchers have previously found that abstract words like “God” and “Satan” have associated directions that incidentally help people to complete both semantic and spatial tasks. People intuitively look upward after reading “God” and downward after reading “Satan”, which has been shown to help orient attention to those spatial locations. Cues for abstract words like these cannot, unlike concrete words, come from perceptual experience.

Some of these associations between words are surprisingly rich. We know the positivity of words like “happy” and “hero”, and entire tomes have been devoted to classes of objects that we group together based on emotional dimensions- as in Lakoff’s provocative Women, Fire, and Dangerous Things, which puts the role of metaphor center stage in human cognition.

One question we can ask is: How did we come to associate some words, such as “God”, with a direction, if we cannot directly perceive the things they refer to? Do language use and perceptual processes affect each other?

In a study recently published in the Psychonomic Bulletin and Review, researchers Goodhew, McGaw and Kidd at The Australian National University were interested in the origin of the spatial properties of abstract words like “God”, “Satan”, and “dream” and “delay”. If language affects embodied expressions, then linguistic behavior should predict visual attention.

To do this, the authors used techniques from computational linguistics that look at the statistical behavior of words. In computational linguistics, we can measure the relative degree of association between two words, which at high levels of association are called collocations. By doing this, Goodhew and colleagues were able to see how much a positive word like “dream” co-occurs with words used to describe a higher physical space, like “up” and “above”. They used a relatively simple method, tallying the number of times phrases like “God above” occurred in a large text database, known as a corpus. The more a word occurs with one particular direction word (say, more “up” than “down”), the more presentation of that word should automatically activate that direction.

Goodhew and colleagues created a set of words that varied in how “up” or “down” they were, as well as how closely associated these words were to “up” and “down” words. Half of the words were associated with each direction based on previous studies (e.g. “God” is up, and “Satan” is down). Half of the words were concrete (e.g. “sky”, “ground”), and half of the words were abstract (“dream”, “delay”). All words had an associated linguistic direction. The authors predicted that words’ linguistic biases would contribute to the conceptual cuing effect.

The authors reasoned that when a word is presented, it will cue the spatial location to which it is associated. Participants’ attention should thus be oriented toward that direction. If a target stimulus is presented in a place that is consistent with the direction that the word cues, the target should be easier to identify, as attention will already be in the right location. Therefore, we should look up when we read “dream” and “sky”, and down when we read “delay” and “ground”, even when it does not help us do the task.

To test this idea, Goodhew and colleagues used a letter identification task. The words were presented at the center of the screen. The word disappeared, and then a “k” or a “l” appeared either above or below the center of the screen and participants had to identify the letter with a key press. Conceptual cuing predicts that words associated with “up” and “down” will cue those directions, thus orienting attention.

Exactly as expected, when the word was conceptually associated with “down”, such as  “ground”, people processed letters shown at the bottom of the screen faster than if the cue word was associated with “up”, like “sky”. We therefore seem to automatically orient our attention in a direction that is tacitly implied by the abstract word we are processing—“God” really is “up there” somewhere, psychologically speaking.

Goodhew and colleagues also tested whether linguistic biases explained this cuing effect. They found that cuing based on linguistic co-occurrence statistics was equally true for abstract words and concrete words. Most surprisingly, linguistic factors (the degree to which a word is associated with “up” or “down”in the statistical sense) were just as good at explaining the cuing effect as the conceptual spatial orientations that words had been assigned in previous studies.

The authors conclude that “Either language use is the cause of the associations between concepts and space, or language use reflects a consequence of an implicit, pre-linguistic mapping arising from our perceptual experiences with objects in the world around us.” We seem to learn that abstract words have spatial orientations from the statistics of our language, which explains how we can have embodied representations of very intangible things.