In contrast to naturalistic studies of spoken language, language comprehension is typically investigated in a more formal experimental setting.
"Click-studies", where a noise is played at the same time as a piece of recorded speech, appear to show that the ability to perceive and process linguistic information is very tolerant of noise. For example, subjects hearing the word "boy" with a click obscuring the centre vowel of the word report hearing the word "boy" with a click either before or after it. Similar tests with larger words and even phrases seem to show that the brain is able to distinguish linguistic meaning from a very noisy environment. For example, the phrases:
Many studies have been carried out into the way in which meaning is stored in our memory. These show that people remember meaning rather than content of sentences: the longer the gap between hearing a sentence and being asked to recall it, the harder it is for a subject to remember the exact words, although the meaning is relatively easily retrieved.
One field of research is related to "centre-embedded" sentences, which are
allowed in English. For example, in "the lion the dog bit
roared", it seems fairly straightforward to parse the meaning, and it
appears that the embedded structure "the dog bit" can be treated as a distinct
unit without getting confused with "the lion roared", i.e.
the lion roared.
the dog bit
However, while there doesn't appear to be any grammatical rule preventing
more embedding (in the same way as adding more and more adjectives in front of
a noun), it doesn't seem at all straightforward to understand the meaning of a
sentence if we just embed one more clause: "the lion the dog the cat scratched
bit roared", which ought to break down like this:
the lion roared
the dog bit
the cat scratched
This might tell us something away that our parser works (e.g. although the
first example suggests a stack, the second doesn't).
There's a fair amount of stuff on "centre-embedded" sentences on the web.
Studies have shown that when asked to react to the sound of a certain phoneme, reaction times are slowed when the phoneme is preceded by an ambiguous word. For example "the men started to drill before they were told" contains the phoneme /b/ straight after an ambiguous word. By substituting "march" for "drill", reaction times drop, suggesting there's some interference due to the extra effort required to parse the awkward word. This effect quickly disappears though: in the phrase "the men started to drill very energetically before they were told", the ambiguity is still present, but by the time the /b/ appears, it seems that the extra effort to process it has been "turned off": reaction times to the /b/ in this case are not significantly different than if there had not been an ambiguous word.