Thursday, May 19, 2011

More discussion of Brocas' area and sentence comprehension

A couple weeks ago Andrea Moro emailed me in connection with the recent critical review of the relation between Broca's area and sentence comprehension published by Corianne Rogalsky and myself. She pointed out a few highly relevant papers that we had unfortunately neglected in our review. An interesting email exchange ensued involving myself, Andrea, and Cornelius Weiller. I'm publishing this exchange with everyone's permission for two reasons. One is point out this important body of work that we did not address (it's worth a serious look) and the other is just because I thought the discussion presented some important issues that the broader language neuroscience community would be interested in and might like to comment on.

Here are the papers Andrea pointed out to us:

Tettamanti M, Rotondi I, Perani D, Scotti G, Fazio F, Cappa SF, Moro A. Syntax without language: neurobiological evidence for cross-domain syntactic computations. Cortex. 2009 Jul-Aug;45(7):825-38.

Tettamanti M, Alkadhi H, Moro A, Perani D, Kollias S, Weniger D. Neural correlates for the acquisition of natural language syntax. Neuroimage. 2002 Oct;17(2):700-9.

Musso M, Moro A, Glauche V, Rijntjes M, Reichenbach J, Büchel C, Weiller C. Broca's area and the language instinct. Nat Neurosci. 2003 Jul;6(7):774-81.


And here is the exchange, starting with my response to Andrea (there were multiple people copied on all these emails and they have been edited down minimally):

----
Greg wrote:
If broca's area is critically involved in hierarchical computations shouldn't patients with lesions there have significant trouble constructing sentences and making grammaticality judgements? As you know, focal lesions to broca's area do not cause broca's aphasia and rather causes only minimal if any language deficits.

----
Andrea wrote:
It seems to me that there is a misunderstanding here: "constructing sentences and making grammaticality judgments" as you say is surely a process that involves recursion but it is also undoubtedly true that there are cases where you can compute meanings with different strategies that do not involve recursion in the relevant sense. Take as a simple example a pure list of words like [lions, frogs, killed]: you can even think of them as in a Venn diagram. There surely is a way to derive a meaning without involving complex nested dependencies, so I would be more careful to jump to the conclusion you suggested. As for the lesions in Broca's areas (and let me just note that anyone in the recipient list of this message is surely much more entitled than me to comment on clinical aspects), it should be highlighted that the syndromic deficits that undergo the common name of "Broca's aphasia" are rather associated to much more extended lesions than those selectively involving BA 44-45. I'm sorry if these poor remarks on my side are not as extended as they should but they would obviously take more time than the one permitted in a simple e-mail.

----
Greg wrote:
Re: broca's area and broca's aphasia you are correct -- the syndrome is associated with a much broader lesion -- and that is precisely my point. When BA 44/45 -- the region implicated in hierarchical processing -- is selectively damaged, it has little effect on expressive or receptive language abilities. This is a problem for the claim I think, no?

----
Cornelius wrote:
Broca is a heterogenous area with different parcellations, each participating in various networks. When we acknowledge this, many questions could be answered.
I am attaching a paragraph from our recent paper in Brain & Language, which I thought Greg, you had reviewed. We will have a meeting in octobre here on processing along dorsal and ventral streams and Broca, hierarchy and recursivity will play a major role there.

A dual stream system has implications for the understanding of aphasic syndromes and recovery from aphasia. A bilateral, left-lateralised, parallel processing system with both a ventral and dorsal connection between temporal and frontal language zones gives ample options for compensatory processes after focal lesions resulting in a variety of active combinations of connections between cortical regions, which differ in a quantitative rather than an absolute manner. This assumption makes the complexity and fuzzy link of aphasic syndromes to focal lesions more understandable.


or as I put it in another version of this manuscript, which went through so many revisions I never had before:

In such a network model, the different cortical regions interact closely, and complex cognitive functions emerge from a context depending It is the selection of the connected regions that determines the function and the domain (see also (Damasio, 1989)(Vadia et al. , 1995)). This means that the different regions are not functioning independently of each other. Thus, the destruction of the interconnection may not result in a solitary failure (e.g.,repetition) but in a completely new phenomenological constellation, as the lesion of the tract affects the functioning of the two regions it connects and other regions, including the right hemisphere, may be operational after the lesion. These assumptions make the complexity and fuzzy link of aphasic syndromes to focal lesions more understandable.

----
Greg wrote:
Thanks for the interesting discussion. I don't disagree with anything that has been said, and these are important points: Broca's area is not unitary and it does not function in isolation but rather part of a complex network. However, I think this misses my point just a bit.

The fact is that damage to ALL of the various subdivisions has relatively mild effects on sentence-level processing and isn't typically associated with full-blown agrammatism, etc. So the fact that there are subdivisions is quite irrelevant.

While it is true that one might account for the lack of a profound deficit following such lesions by appealing generally to the broader network distribution, one still needs a theory of this network that can explain the findings. As far as I'm aware, there is no such theory. What I am seeing in the literature, is the claim that Broca's area (or BA 44 more specifically) is specialized for hierarchical processing. These activations seem to be highly focal suggesting that this region is a critical node. If it is critical, then damage to that region should produce a substantial change in the ability, yet it does not. My point is that this needs to be accounted for in our theories.

Part of our goal with the Broca's area review was to push the field to look at the whole constellation of data, including lesion work, which it has not done as well as it could.

----
Andrea wrote:
Not at all: your papers indeed raises a fundamental question, and the immediate debate that it raised witnesses its intrinsic value. The only aspect that worries me (but it's not at all related to this paper of yours) is that there is a lack of knowledge of the real hierarchical (and recursive) processes that syntax is build upon. For example, I have often seen authors confusing hiearchical with recursive. All recursive structures are hierarchical but not viceversa, witness syllable structures! Another typical confusion which many papers manifest is the one between recursion and nested dependencies (which in a sense are the only recursive structures that are relevant to evaluate syntactic dependencies). Needless to say, again, your own paper as well as the comments by Cornelius and Stefano have been at least for me the source of interesting thoughts.

----
Cornelius wrote:
Dear Greg,

You are basically requiring the integration of a holistic with a localist view, actually quite similar as Wernicke did. We promote two prominent ways to introduce hierarchy in networks leading to a “Concept of critical lesions“ (Weiller et al 2010):
the importance of a network node may be determined by the number of connections, active during a given task (e.g., IFG and MTG in sentence comprehension Saur et al 2010) within a network, the necessity of a given brain region may be determined by a basal and domain general function, e.g. analysis of hierarchical structure in Broca‘s area (e.g.; Musso et al 2003)

Still, predictions should be possible. If the critical, domain general regions for hierarchical structure processing in IFG and its appropriate connections are destroyed, agrammatism should be expected. However, the lesion will have an influence on the other players in the network, which are changed in their behaviour, resulting in a complete new phenomenology. We will also have to take into account the individual variability of anatomy and capacity for reorganisation. There are lesions of Broca‘s area without agrammatism. However, in the neuropsychological literature we are not aware of reports about isolated and complete lesions of all parts of Broca‘s area. Therefore, complementary functions may be taken over by the remaining parts of Broca‘s area and its connections (e.g.; within the dual loop model)

Weiller C, Saur D (2010) Recovery from Aphasia: lessons from imaging studies. In: Cramer S. and Nudo R. Brain Repair after Stroke. Cambridge University Press, New York

Saur D, Schelter B, Schnell S, Kratochvil D, Küpper H, Kellmeyer P, Kümmerer D, Klöppel S, Glauche V, Lange R, Mader W, Feess D, Timmer J, Weiller C (2010) Combining functional and anatomical connectivity revels brain networks for auditory comprehension. NeuroImage 49: 3187-3197.

Musso M, Moro A, Glauche V, Rijntjes M, Reichenbach J, Büchel C, Weiller C (2003) The role of Broca’s area and the language instinct. Nature Neuroscience 6:774-782

2 comments:

VilemKodytek said...

If the auditory recognition doesn't include phonemes (Vaden et al.), I wonder about how does the reading of alphabetic texts work. Does it mean that Broca's area is necessary for reading?

Greg Hickok said...

Good question. I haven't looked into this very thoroughly, but yes, I think that may be true to some extent. Reading activates Broca's area much more reliably that spoken language. This may be because reading -- at least one reading route -- involves grapheme-to-phoneme correspondence. If phonemes are represented not in auditory cortex but in motor-related areas, this makes some sense.