How do we process language as we see or hear it? You can predict the future (a little bit)!

Language Processing Models: Top-Down Versus Bottom-Up

Bottom-up pathways rely on data as it comes in. These models process information by building up a larger meaning step-by-step from basic units, kind of like climbing up a staircase. For example, if you used a bottom-up method while reading, your eyes would transmit each individual letter to your brain, and your brain would tie together all of this information to construct the appropriate words and sentences, without any guess of what comes next.  This is useful when you don’t know what to expect, and for updating your knowledge as you continually have new experiences. 

a black and white image of a staircase with a filigree banister
Bottom-up processing is like walking up stairs: you can only get to the next step by building on the stairs you’ve already climbed.

Unlike bottom-up processing, top-down pathways don’t wait for the most detailed information. Once you have a lot of knowledge about the type of information you’re processing, such as how words are commonly strung together in your language, you can make predictions about what will come next based on the context.

If you told someone you were going to take up a beekeeping hobby, but you didn’t have a suit, and they said to you “Oh no, please don’t g-” your brain will probably already have filled in the end of the sentence (“get stung!”). You knew the topic of conversation (bees), the type of phrase that usually comes next in this sentence structure (a verb phrase or predicate), and the phonetic information provided by the speaker preparing to say the next sounds ([gɛ]), and all that information helped you predict the most likely end of the sentence.

a honeycomb tray covered in bees being held by someone not wearing gloves. A speech bubble says "Don't get stung"

When those predictions are correct (which is actually most of the time!), your brain is already expecting the information, so comprehension is very easy and fast. These pathways use more brain power than bottom-up pathways, and are usually only possible once someone knows a lot about their language (so this type of processing takes some time to arise for babies learning their first languages!). This type of processing also helps when you’re talking to someone in a noisy room or with a different accent– it helps you “fill in” the gaps of acoustic information that you didn’t actually hear in real time, or that sounded different than something you would expect to hear the context of your conversation.

The lowest level of information in language processing is the phoneme or chereme, all the way up to higher levels where you can work out the meaning of what someone said. Kids’ ability to do top-down processing increases throughout language development as they build up prior experience and start to expect common patterns. Some studies show stages of development where kids start to do predictive processing, but have trouble changing their guess when it turns out they’re wrong. This type of learning is very cool for scientists because we can see kids’ skills developing over time!

As our language processing systems develop, we learn how to use a combination of both mechanisms to process incoming information in an “interactive” model of language processing. We can see both types of processing working together through a special type of sentence that linguists call “garden path” sentences.  Read the following sentence:

The old man the boat.

It’s probably pretty confusing at first, right? Your brain started reading [The old man] and started to build a very common syntactic form top-down (meaning you assumed that ‘old’ was an adjective describing the noun ‘man’); but then, it gets to another noun phrase [the boat] and you realize that the initial structure was wrong. In this case, “the old” is a collective noun referring to a group of elderly people, and “man” is used as a verb (meaning to stand by or be part of the boat staff). 

Interactive models of language processing suggest that top-down and bottom-up processes work together simultaneously to process language, so you can always correctly incorporate higher or lower level information. Higher level information allows lower levels to adjust to the given context, making information processing more accurate. For example, if you hear the sentence “you’re on the wrong route/root,” your bottom-up processing will recognize the two possible meanings of the final word: a course for traveling and the part of a plant that gets water from soil. Top-down processing will help choose which is correct depending on the context, whether it’s a group of kids playing around the base of a tree or people in the car on a road trip.

Brain-mapping studies show that both types of processing occur in the early and late stages of language processing.  This suggests teamwork between top-down and bottom-up processing for people making sense of all of the language around them!

 

Resources

Bitan, T., Cheon, J., Lu, D., Burman, D. D., & Booth, J. R. (2009). Developmental increase in top-down and bottom-up processing in a phonological task: an effective connectivity, fMRI study. Journal of cognitive neuroscience, 21(6), 1135–1145. https://doi.org/10.1162/jocn.2009.21065

Field, J. (1999). “Bottom-up’ and ‘top-down.” 338-339.

Jackendoff, R. (2013). A parallel architecture model of language processing. In The Oxford Handbook of Cognitive Neuroscience, Volume 1.

McClelland, J. L. (1987). The case for interactionism in language processing. CARNEGIE-MELLON UNIV PITTSBURGH PA DEPT OF PSYCHOLOGY.

Moore, D. R. (2012). Listening difficulties in children: Bottom-up and top-down contributions. Journal of communication disorders, 45(6), 411-418.

Noesselt, T., Shah, N. J., & Jäncke, L. (2003). Top-down and bottom-up modulation of language related areas–an fMRI study. BMC neuroscience, 4(1), 1-12.

Osada, N. (2001). What Strategy Do Less Proficient Learners Employ in Listening Comprehension?: A Reappraisal of Bottom-Up and Top-Down Processing. Journal of Pan-Pacific Association of Applied Linguistics, 5(1), 73-90.

Shuai, L., & Gong, T. (2014). Temporal relation between top-down and bottom-up processing in lexical tone perception. Frontiers in behavioral neuroscience, 8, 97.



Taylor is laughing and looking at the camera. She wears a grey and white t-shirt, has light skin, and curly light brown hair. The clue blue sky and some grass are visible in the background.

Taylor Horowitz

Author

Taylor is a recent graduate of Duke, and is now a Clinical Research Coordinator at Cohen Children’s Hospital. After Duke, she hopes to attend medical school. Taylor is interested in learning about how children’s environments influence their ambitions and goal-setting, both academically and socially. 

Elika Bergelson

PRincipal Investigator