Of Minds and Language (63 page)

Angela D. Friederici

Let me begin with a little anecdote. When I came to MIT in 1979, I was full of the energy and proud of the data derived from my Ph.D. research. Very early on, actually during my first week at MIT, I was able to take part in a workshop and there I came to sit at a table next to a person whom I didn't know, but whom I told all about my wonderful work in reading and writing, and this person said to me, “Why do you think this is interesting?” [laughter] And you guess who that person was. It was Noam Chomsky. As a result of this my entire career has focused on auditory language processing, and so in today's talk I will discuss the brain basis of auditory language comprehension.

In a lecture like this, I think we have to start from scratch, namely with Paul Broca (1865). As you all know, in 1865 he discovered a patient who was unable to produce language; he was able to produce only a single syllable, the syllable “tan.” This person was well described by Broca, who later on after the patient had died was able to look at the brain, and what he found was a lesion in the inferior frontal gyrus (IFG) of the left hemisphere. This lesion was already large when looked at from the outside, but about a hundred years later the patient's brain was found on a shelf in one of the anatomy institutes in Paris, accurately labeled M. Leborgne (the name of the patient described by Paul Broca). At that time we had the first structural imaging techniques, such as computer tomography (CT), and thus Leborgne's brain was put into a CT scanner. Interestingly enough, Broca had already predicted that the lesioned area should be pretty large. He was able to do so by the following means. He had put a little metal plate on the brain and knocked on the metal plate. And from listening to how the brain tissue responded to his knocking, he concluded the brain lesion must be very deep. He described that, and the CT provided the proof (
Fig. 22.1
).

Fig. 22.1. Brain of the patient described by Paul Broca in 1865.

The black region on the CT scan is the lesioned area, and it was very large – much larger than what we call Broca's area today. Broca's area today is defined to include Brodmann area (BA) 44 and BA 45 (
Fig. 22.2
).

A couple of years later, Carl Wernicke (1874) saw and described six patients who seemed to have kind of a reverse language pattern. The deficiency of these patients was one of comprehension. They could not understand simple commands or sentences, but they had a fluent language output which, however, was without much content. In those times, such patients were often not considered as having a language problem but as having a thought problem. Wernicke described the lesions of the patients as being located in the superior temporal gyrus (STG). Later this region was called Wernicke's area (
Fig. 22.2
) and taken to be relevant for language comprehension. Nowadays we have to revise this
classical neuroanatomical model, separating comprehension and production into Wernicke's area and Broca's area respectively, because we know that Broca's area is not only involved in production, but also in language comprehension. I will specify the revised model, which is based on recent brain imaging methods, in this paper.

Fig. 22.2. Left hemisphere of the human brain. Numbers indicate areas which were described as cytoarchitectonically different by Korbian Brodmann in 1909.

Fig. 22.3. Model of auditory language comprehension. For details see text.

Source
: Friederici and Alter 2004

The research questions that we can address with the new neuroscientific imaging methods are the following. We can ask which brain areas support sentence processing, and particularly syntactic, semantic and prosodic processes. To answer these questions we use functional magnetic resonance imaging (fMRI). Moreover, we also have the possibility to look into the dynamics of brain activation, namely with methods which enable us to trace the time course of brain activation. These are electroencephalography (EEG) and mag-netoencephalography (MEG). Although the spatial resolution in EEG and MEG methods is restricted, these methods can tell us something about the specific temporal relation between the different processes. What we need, however, when we want to look at language-related brain activation in a systematic manner, is a functional model. The model in
Fig. 22.3
is a coarse model – no doubt about that – but it has some special features which I would like to point out. There are two pathways, one on the left side, one on the right. Later on we will see that the processes sketched in the pathway on the left side are mainly performed by the left hemisphere (LH), and those of the pathway on the right side are mainly performed by the right hemisphere (RH).
¹
So what exactly are the functions of these hemispheres?

Let us first consider the left-hemispheric processes. The idea here, and it is a strong prediction in the model,
²
is that there are separate phases in processing syntactic and semantic information, and that the phases, following Lynn Frazier and Janet Fodor (1978), are sequential. In the first stage word category information is accessed and local phrase structures are built. This first processing stage is totally independent from semantic information. And then only in a second stage you access lexical-semantic information and assign thematic roles. Certainly, there are other psycholinguistic models which assume a strong interaction between these two components and at each moment in time. The model proposed here is a strong model and we can see how far we can hold up the hypothesis that the two processes are really serial. All these processes work incrementally. The system does not have to parse the entire sentence before entering the next processing stage, but can proceed in a cascade. Then at some final integration stage the system has to map the output from the two other stages to achieve comprehension.

With respect to the RH, there is the suggestion that prosodic information is processed in the RH. This holds, without any doubt, for emotional prosody, but here the focus is on linguistic prosody. Pitch information certainly provides information about intonational phrasing and also about accentuation. Today I will mainly talk about intonational phrasing and I will also discuss how intonational phrasing and syntactic phrasing go together.

Let's first concentrate on the processes assumed to be located in the LH, and see which brain areas support semantic and syntactic processes. We will do so by looking at a couple of studies using functional magnetic resonance imaging (fMRI). In our institute we usually try to scan the entire brain in order not to miss important activations in areas of the brain not predicted to respond to language. In a first experiment (Friederici et al. 2003a), we thought that one way to disentangle the semantic and syntactic information which usually comes together in a sentence would be to work with a violation paradigm. That is, we presented semantically incorrect sentences, for example sentences containing a violation of selection restrictions, such as:

(1)
*Das Lineal wurde gefiittert
.

‘The ruler was fed.'

For the syntactic part, we presented syntactically incorrect sentences, for example:

(2)
*Die Ente wurde im gefiittert
.

‘The duck was in the fed.'

This sentence is incorrect for the following reason. In a prepositional phrase in German, the preposition
im
, which already carries a case marker (in translation:
in-the
), requires a noun or adjective + noun combination to follow. What the subjects perceive, however, is a verb; that is, we have a violation of the word category and the question is how would the brain react to this violation?

We also included correct sentences where the prepositional phrase was fully present, e.g.:

(3)
Die Kub wurde im Stall gefiittert
.

‘The cow was in-the barn fed.'

The syntactic violation stimuli were manipulated, in order to avoid acoustic cues of “incorrectness.” As speakers invariably lengthen the preposition in such incorrect sentences, providing an acoustic cue of “incorrectness” in non-spliced sentences, incorrect sentences were cross-spliced taking the preposition from a correct sentence containing a full prepositional phrase.

When comparing the semantically incorrect to the correct condition (
Fig. 22.4A
) we found a significant difference in the posterior and middle portions of the superior temporal gyrus (STG), but not in its anterior portion. For the syntactic violation condition (
Fig. 22.4B
) we found a clear difference in the anterior portion of the STG but also in the posterior portion, and to some extent in the middle portion. Thus what really stands out for the syntactic incorrect condition is the difference of the anterior portion of the STG.

A particular area that was predicted to also be activated when dealing with local dependencies is the frontal operculum. When considering a slice which covers the frontal operculum we found this area to be significantly more active for the syntactically incorrect condition than for the correct condition.

From this, and a number of other studies in the literature, we can define two different networks. The first, the semantic one, which comprises the posterior and middle portions of the STG, and, under some conditions, also activation in the inferior frontal gyrus, that is BA 45 and BA 47. This latter region only comes into play when strategic semantic processes are required, that is, when asked to categorize words into particular semantic categories (Fiez et al. 1995; Thompson-Schill et al. 1997). During online sentence processing, this activation is seldom seen. For the syntactic processes, the network consists of the anterior portion of the left STG and the frontal operculum right next to, but not identical with, Broca's area in the inferior frontal gyrus (IFG). The posterior portion of the STG, which is seen to be active during semantic and syntactic processes, may be considered to be a region where these information types are integrated.