Seeing not just any evil: Eye Movements, Typos, and and Autocorrects

I find that the older I get, the more I “accidentally a word.” And that’s before we get to autocorrect. Sometimes, word processor errors coupled with collaborative editing on a document mean that someone types the same word in their highlighted text as occurs in the next section of the text. Or, occasionally, people will write the same words multiple times (e.g. “the the” or “same same”), showing either a perseveration error, or an anticipation error. Typos abound: by my informal count of words on the Twitter account of the New New York Times, nearly 12% of “new words” are actually just typos. In some domains, such as texting or social media, the incidence of typos or missing words may be even higher.

Reading is one of our most sophisticated motor skills, with fine eye movements allowing us to interpret written symbols, and turn them into a meaning that we can understand. Consider for example this video of where eyes look as we’re reading a sentence:

Reading is sensitive to both low-level information, such as the blurriness of the text or the similarity of a written symbol with others, in addition to high-level information, such as expectations we build up about where a story is going. As you can see in the above video, our eyes tend to focus on “content” words, or words that carry meaning, such as “dog”, “big”, “metal”, “muscles”, and so on.

Words that we focus on, or fixate, are words that are seen in the fovea, the part of the eye with the greatest acuity. One reason this might happen is because little words like “the” can be seen without directly looking at them because they are small enough to be processed just outside the fovea – in the parafovea. Readers also tend to skip over words like “the” that are repeated next to each other in the text. In a classic study asking participants to count letters (e.g., “e”) in a short text, the count will typically be closer to the number of “e”s in content words than the total count including the “e”s in function words. This may again be because function words tend to be skipped, while content words tend to be fixated.

So just how do readers handle a seemingly endless stream of typos? If we accidentally a word, or accidentally add a a word, do readers tend to notice? What if we accidentally used the same same word twice? That is, are some errors easier to spot than others?

Researchers Adrian Staub, Sophia Dodge, and Andrew Cohen tackled this question in a recent paper in Psychonomic Bulletin and Review. Their specific question aimed at understanding what factors influence readers’ ability to find duplicate or missing words, and whether this is due to eyes simply skipping over words or whether it even occurs when our eyes land on those words. If we still fail to notice repeated words when we read them, this suggests that our tendency to not notice function word errors is the result of a higher-order cognitive process than just eye movement control.

Staub and colleagues compared simple grammatical sentences such as “Amanda jumped off the swing and landed on her feet” to ones with subtle changes:

  1. “Repeated ‘The’”: Amanda jumped off the the swing and landed on her feet.
  2. “Repeated Noun”: Amanda jumped off the swing swing and landed on her feet.
  3. “Omitted ‘The’”: Amanda jumped off [] swing and landed on her feet.

Participants read sentences like these and then had to answer comprehension questions involving a binary choice. For critical items such as the ones above, participants were asked “Was there anything wrong with that sentence?” and had the binary choices of YES and NO. For filler items like, “Dan went outside. He discovered that it was much colder outside than inside.”, participants answered different questions such as “What was colder?” with distinct responses (e.g., INSIDE/OUTSIDE).

Because of the binary-judgment nature of the critical trials, participants could either produce hits (identifying an error when there was one) or false alarms (stating there was an error when there was none). In the normal grammatical condition, participants answered 96.1% of the yes/no error questions correctly. In the repeated noun condition, participants answered 90.2% correctly, showing that they only failed to notice errors 9.8% of the time. By contrast, when “the” was repeated, they missed it 54.2% of the time. When “the” was missing entirely, participants missed the omission 32.5% of the time.

Do these mistakes in identifying repeated or omitted words mean that participants were skipping those words, leading to errors? Staub and colleagues therefore conducted an analysis of participants’ errors depending on where their eyes fixated when they were reading these sentences. Staub and colleagues were specifically interested in whether participants’ eyes fixated a word any time before they moved on to a later part of the sentence.

In normal grammatical sentences, readers typically did not make comprehension errors. When looking at participants’ eye movements, Staub and colleagues found an interesting pattern of results. In grammatical sentences, participants read the word “the” 49.5% of the time and the noun (e.g. “swing”) 75.8% of the time. For ungrammatical sentences, there were two types containing repeats: the “Repeated ‘The’” (RT) and “Repeated Noun” (RN) conditions. In RT conditions, if participants skipped the first “the”, they read the second one 78% of the time. But, if they read the first “the”, they would fixate the second one only 38% of the time. This proportion is very similar to any chances of fixating a word “the” if the previous word was fixated—that is, in other stimuli, participants would fixate any “the” about 33-38% of the time. If participants read both “the”s, they detected the repetition error 66% of the time. When they did not read both or either “the”, error rates were worse, with participants only detecting the repeat 34-45% of the time.

When nouns were repeated, participants would usually read the first noun, 75.8% of the time, which was comparable to 72.3% in the normal sentences. If the first noun was skipped, participants read the second one 92.8% of the time, but if they had fixated the first one, the probability of looking at the second one dropped to 75%. Irrespective of whether someone fixated the first noun only, the second only, or both, accuracy in detecting the duplicate word was roughly equal (about 90%).

The pattern of results here shows a dissociation between content and function words. Participants almost always noticed when content words like “swing” were repeated, but very often did not notice a repeated “the.” Readers’ insensitivity to the repetitions of “the” could be because the information the eye sends to the brain is noisy. For example, during eye movements, known as saccades, no (or little) visual information is sent, and so it must be reinterpreted afterward – the brain interpolates.

Another possibility is that because function words like “the” fit into the parafovea (see Figure below), there may be some uncertainty that it was there at all. This is especially plausible because eye movements often “automatically” skip over short words when they fit with the context.

Source: Schotter et al. 2011. 

While the mechanisms that are involved in skipping words and identifying mistakes are still to be fleshed out, odds are that some kinds of mistakes will go relatively unnoticed. So, if you forgot to correct your text, or if you even accidentally a word in your term paper, the mistakes may fail to register. On the flip side, this also means that catching your own mistakes will still be hard, but now you know not to sweat it.

Psychonomic Society article featured in this post:

Staub, A., Dodge, S., & Cohen, A. L. (2019). Failure to detect function word repetitions and omissions in reading: Are eye movements to blame? Psychonomic Bulletin & Review, 26, 340-346. doi: 10.3758/s13423-018-1492-z

The Psychonomic Society (Society) is providing information in the Featured Content section of its website as a benefit and service in furtherance of the Society’s nonprofit and tax-exempt status. The Society does not exert editorial control over such materials, and any opinions expressed in the Featured Content articles are solely those of the individual authors and do not necessarily reflect the opinions or policies of the Society. The Society does not guarantee the accuracy of the content contained in the Featured Content portion of the website and specifically disclaims any and all liability for any claims or damages that result from reliance on such content by third parties.

You may also like