Thursday, November 5, 2009

Why "where" cannot be a sensory processing stream

There is debate about the nature of the dorsal auditory processing stream. Some folks, Josef Rauschecker in particular, argue for a dorsal "where" stream, whereas others, Hickok & Poeppel and Warren et al., argue for a sensory-motor integration (sometimes called "how") stream. Here's why the "where" hypothesis can't, in principle, be right.

Spatial information associated with an auditory signal is a stimulus feature much like pitch. We don't talk about a "pitch stream" however. Why not? Because pitch (frequency) is just a cue for any number of processing goals. Pitch information can cue phonemic identity, speaker voice, auditory stream segregation ... even sensory-motor goals (humming back a tone). Spatial cues are no different in that they can cue explicit location judgments, auditory stream segregation, and any number of sensory-motor processes (head movements, saccades, locomotion toward or away from a source).

Processing streams, I'm suggesting, are defined by goals or tasks -- what the information is used for -- not by stimulus features. Sensory-motor integration for vocal tract actions defines a goal -- control of the vocal tract -- and therefore is a viable candidate for a processing stream. Identifying the meaning of an auditory object is also a goal and a good candidate for a processing stream. Stimulus features, like pitch or location are not goals, they are just cues that can be used within various task-driven processing streams.

Of course, this doesn't imply that there isn't a specialized location processing system in the brain that uses interaural time and level differences to compute spatial information. Almost for sure there is (my guess is that it's subcortical), just like there is a system that processes pitch using frequency information. But we shouldn't confuse a specialized feature processing system (area) with a cortical processing stream as the notion "stream" is typically used.

Which reminds me. It is probably time to redefine the notion of a processing "stream". In particular, I think the dorsal-ventral distinction is getting tired and has now outlived its usefulness. I'll expand in a later post...


Hickok, G., & Poeppel, D. (2000). Towards a functional neuroanatomy of speech perception Trends in Cognitive Sciences, 4 (4), 131-138 DOI: 10.1016/S1364-6613(00)01463-7

Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing Nature Reviews Neuroscience, 8 (5), 393-402 DOI: 10.1038/nrn2113

Rauschecker JP, & Scott SK (2009). Maps and streams in the auditory cortex: nonhuman primates illuminate human speech processing. Nature neuroscience, 12 (6), 718-24 PMID: 19471271

Warren JE, Wise RJ, & Warren JD (2005). Sounds do-able: auditory-motor transformations and the posterior temporal plane. Trends in neurosciences, 28 (12), 636-43 PMID: 16216346


Kevin H said...

Why wouldn't determing where a sound is coming from be a viable goal? While some aspects of location are certainly easily handled by the brainstem, so are some aspects of object identification. Determing where and object is in space is much more of a issue than simple ITD/ILD. There is almost certainly some input from AM sensitives cells to detect auditory motion, large visual input, and heavy interaction with spatial attention. None of which can be adequately handled subcortically.

I don't really doubt your findings for spt, but it is a bridge too far to extrapolate those findings to everything auditory related dorsal of HG.

Greg Hickok said...

Hi Kevin,

You make some good points. Let me ask you this before I answer: why are you not defending a "pitch stream"? Isn't determining what pitch a sound is a viable goal?

Kevin H said...

Pitch is certainly an aspect of 'what'. I guess it all comes down to how parsimonious you want to be. Each neural assemblage has it's own unique role, and therefore there will be a task that is best answered by that particular group of cells, but we don't give them all their own 'stream'. Stream just becomes a useful shorthand for a number of different assemblages that have similar or overlapping goals.

I think you can certainly take the generally useful idea of streams and make it into useless dogma. For example, take AM. AM is important in both determining what you are hearing, and determining where it is/will be. A silly dogmatic approach to streams would say that you had to have two separate AM processing regions, one that was devoted to 'what' processing and one that is all 'where'. A much more pragmatic vision of streams is that these processes are all somewhat overlapping and complex, and the brain will try to solve it all in the most efficient system possible, which won't fit perfectly into the boxes we want it to.

In that same sense, I can certainly see sensory-motor areas fitting in quite nicely. Obviously, ever part of the processing stream needs to get back to a motor system eventually, and some systems will need to make that leap sooner rather than later. There are a number of reason that this sensory-motor system might be biased to be caudal of HG. Being closer to motor cortex improves reaction times slightly, and it's motor functions may overlap with other motor functions such as orientating to a sound that have a lot to do with the spatial processing happening in nearby regions.

So, I guess I'm trying to get at this: If someone tries to tell you that a given area CAN'T have a certain function because everything in that general area is devoted to particular 'stream' they are obviously wrong. But, I think your playing into their game if you simply try to redefine the stream. Whatever your definition is, there will be countless outliers. I'd much rather look at streams as a general organizing principle that probably has a lot more to do with anatomical efficiency than serial information flow.

As to what what to label this nebulous, imperfect 'stream'. I'd say that it still looks like a bunch of caudal auditory cortex is primarily devoted to spatial tasks, so I'd still plan on calling it a spatial stream.

Greg Hickok said...

I think you were on the right track when you were talking about dogmatic applications of the notion of a "stream", but you went foul when you stuck to the spatial stream dogma. :-)

Pitch: you say pitch is certainly an aspect of 'what'. But this is not always true. I can play you a sequence of pitches and you can use that sensory information to guide your own vocal tract actions to reproduce that sequence via humming. This is a sensory-motor task and therefore part of the "dorsal stream" (this task also activates Spt very nicely). "Pitch processing" is not a stream because there is no one goal of pitch processing; it depends on the task. The same holds of "spatial processing", what stream it enters into depends completely on the task.

Sensory-motor integration for the vocal tract doesn't suffer this problem because vocal tract control is not an acoustic feature, it is a goal. Accessing a semantic representation is also a goal not an acoustic feature. These are "streams".

Think of it this way: on one level you have a sensory system that analyzes sensory input into various bits of information (pitch, location, etc.). This sensory system then interacts with the rest of the brain to accomplish tasks. The *interaction* of the sensory system with rest-of-brain systems is what defines the streams and any bit of sensory information that is useful for a given task will enter into that stream. In this way of thinking, streams are not really part of the sensory systems (i.e., the concept of an "auditory" or "visual" stream is vacuous), they are the ways sensory systems interact with other neural systems.

Now, in your first comment you said, "wouldn't determining where a sound is coming from be a viable goal?" Sure. But "determining" for what purpose? To control an orienting response? (Dorsal sensory-motor stream.) To separate out one sound source from another? (Ventral stream?) To make decision as to whether the location of that sound source is the same as a previously presented sound source? (Who knows what frontal problem solving stream.) You get the point.

Also, in your first comment you suggested I was extrapolating Spt to be THE DORSAL STREAM. Although it may sound like this from what I write, I don't actually believe it. I think Spt is just the interface between the auditory system and the vocal tract motor system. I think there are other interfaces with say head movement or ocular control systems, and musicians probably have a very nice auditory-manual control system. I.e., lot's of sensory-motor "dorsal streams" exist (note I said 'sensory' not auditory-motor because any relevant sensory information can enter the stream).

Finally, you say that a bunch of caudal auditory cortex is primarily devoted to spatial tasks. I'm not at all convinced this is true. Nobody has actually manipulated the task to see whether the "spatial" activation moves. One thing we have done however is to show that a non-spatial manipulation (adding multiple sound sources) activates caudal "spatial" areas just as much as moving a sound source around (Smith et al. JoCN in press). This suggests that it is not spatial in any obvious way.

Kevin H said...

Interesting... One fine point: I don't think that caudal AC is devoted to spatial tasks, but rather computations that are required for a lot of spatial tasks.

I also think that your emphasis on tasks is a bit too dogmatic. If we needed to process pitch to say, determine if another singer was off key, or to help separate out two voices, would that pitch be processed in a wholy different region (rostral 'what' stream) vs if we were processing our own pitch for vocal corrections (caudal 'sens-motor' stream)? That seems like a very wasteful setup.

Your JoCN article is certainly interesting, but your taking it a bit too far. Use the same logic on early PP/PT distinction papers that had relatively simple pitch/space tasks. Each of those tasks had a pretty straight forward motor component, so would it be fair to say the regions they found couldn't have a sensory motor component? No.

In an aside, I imagine that if you did a SVM analysis similar to what you've done for your SPT paper on the dataset from JoCN, and I bet you'd be able to pull apart some different patterns for movement/number. But I wouldn't concede the above if that didn't pan out =).

Greg Hickok said...

The point is that maybe the "spatial" activations we are seeing in caudal AC are not the basic spatial computations but the higher-order use of that information. Based on the fact that these "spatial areas" seem to respond equally well to auditory object manipulations, maybe caudal AC is where auditory stream segregation is happening. Yes, it is possible that with MVPA we might be able to find spatial selective responses in human AC but to date there is no evidence for this.

Passive listening to pitch vs. spatial variation indeed activates different regions. But just because something responds to stimulus x doesn't mean that it is the "x area". Maybe the caudal AC "spatial" response is really a system that is using spatial information to identify auditory objects. Or maybe it is active suppression of a sensory-motor system that wants to track the signal (subjects are usually told not to move their eyes and definitely not their heads)!

I personally think that it is still an open question whether caudal AC is a spatial computation *area* -- we need more research to confirm one way or the other. However, I am pretty darn sure it is not, and in fact cannot be, a "stream".