Phonemic Restoration
Another phenomenon that shows top-down influences of word knowledge on the perception of speech is called the <b>phonemic restoration effect</b>, first reported by Warren (1970). To get a sense of the effect, listen to the following recording: (this link plays a recording of “The student read the linguistics textbook online.”) If you listened to the recording, you might have noticed that there is a cough over part of the sentence. Ask yourself, on which sound did the cough occur? Could you nonetheless hear that all the sounds in the sentence are there?
If you guessed that the cough occurred at the first /s/ in ‘linguistics,’ you are right. But the science predicts that many people would not be able to report the position correctly. And did you hear that all the sounds were there despite the cough disruption? In fact, in the recording, that first /s/ in linguistics, is missing in the sentence. Using editing software, I replaced the /s/ with silence and then added the cough in that space. But even though there is no /s/ physically there in the acoustic signal, many would report having heard it. This is because our brains can ‘fill in’ the missing sound using what it knows about speech sounds, words, and which words are likely in a particular sentence. In the case of ‘linguistics’, there aren’t any other words you can make by replacing the first /s/ (e.g., ‘linguiftics’ and ‘linguibtics’ aren’t words of English), so we would predict this is a pretty easy sound to ‘fill in’.
This example was made to be somewhat like the original study by Warren. Warren presented undergraduate student participants with recordings of sentences like “The state governors met with their respective legislatures convening in the capital city,” but the recordings were altered so that one of the speech sounds, like the first /s/ in ‘legislatures’ along with any parts of the surrounding sounds that would give acoustic clues that this was an /s/, was deleted to make about 120ms of silence in the middle of the word. This silent period was then replaced by a cough sound. The participants reported having heard all of the sounds in the sentence even though one had been removed, and they were also unable to accurately report where in the sentence the cough had occurred. This is another example of listeners reporting a sound that is not really there in the speech input based on their knowledge of a language, or in other words based on top-down information.
In our ‘legislatures’ just like ‘linguistics,’ there is no other plausible sound that listeners could report instead of /s/. No other sound would form a real word of English, and besides, the word *legislatures* is highly supported by the context of the sentence (assuming that the participants know something about the American political system). What happens when the removed sound creates *ambiguity*, meaning that there is more than one sound that could plausibly be inserted, and therefore more than one word that could be perceived?
This question was part of the investigation conducted by Groppe and colleagues in 2010. In two experiments, the researchers had participants listen to sentences with phonemes replaced with sounds (in this case a tone rather than a cough), for example *He had fallen while climbing a __ountain*, where the missing phoneme could be understood to be /f/, making *fountain*, or /m/, making *mountain*. The sentences were also preceded by a context sentence that was either helpful in deciding which word was intended (*Victor had to get airlifted out of the Rockies*) or neutral (*Victor had to go to the hospital*). The results showed that having the helpful context sentence led listeners to indicate that they heard *mountain* more often than they did with a neutral sentence. Groppe et al. also examined ERPs to the critical words with a phoneme replaced, and found that the N400 response had a smaller amplitude when the word was preceded by the helpful context. This means that listeners interpreted the word with the tone as more expected (*mountain*) when it was within the helpful context, even though acoustically it could have been a word that was quite unexpected (*fountain*).
The findings discussed here have examined vocal languages that are transmitted via sound. But one might ask whether there is any equivalent to restoration effects in signed languages. One study by Schultz-Westre (1985) presented American Sign Language (ASL) signers with videotaped ASL sentences that were edited to replace part of a sign with a blank white screen. Similar to what was found for vocal languages, ASL signer participants were not very good at reporting the location of the disruption in the sentence. This suggests that they had ‘filled in’ the missing material.
The finding of phonemic restoration shows that our minds can compensate for small disruptions in what we hear or see based on our linguistic knowledge and perhaps our knowledge of how the world works and what scenarios are more typical.
References
Groppe, D. M., Choi, M., Huang, T, Schilz, J., Topkins, B., Urback, T.P., & Kutas, M. (2010). The phonemic restoration effect reveals pre-N400 effect of supportive sentence context in speech perception. Brain Research 1361: 54-66
Schultz-Westre, C. J. (1985). A visual analog of phonemic restorations: Sign restoration in American Sign Language. Ph.D. Dissertation, The University of Wisconsin-Milwaukee.
Warren , R. M. (1970). Perceptual restoration of missing speech sounds. Science , 167 : 392 – 393 .
