|
Informational Masking: What is it?
Bill Yost, Parmly Hearing Institute
|
|
In the spirit of the request I received from Barb to provide a provocative statement that might generate discussion, let me start by saying that I never did like Chuck Watson’s original use of informational masking in his work concerning uncertainty and discrimination of tones in his ten-tone, word-length patterns (for clarity, I like Chuck, a lot; just not informational masking). I never thought he was measuring masking nor did he actually measure information. That being said, he provided a valuable contribution in showing that there were two classes of variables that controlled discrimination performance; variables related directly to the stimulus (i.e., how were the oscillator, electronic switch, and attenuators set) and variables related to the way in which the stimuli (whatever they were) were presented over trials. He and today’s literature calls the first energetic masking and the second informational masking or perhaps informational masking in the uncertainty case [IM(U)].Thus, masking is a sum of the two. What was really good about this work was that one variable could be varied entirely independent of the other, so there was no chance of confounding the two.
Let me digress for a brief review of some history. First, I think that Chuck’s recent paper in Acta Acoustica (2005) is a must read, if for no other reason than to get the history of the study of informational masking correct in order to avoid some of the revisionist accounts I have read recently. That is, Watson introduced the term as we use it today based on its use by Irv Pollack in a different context. Carhart (1969) studied speech-on-speech interference and commented on the perceptual masking of speech-on-speech as compared to speech-in-noise, and one can see similarities between the current use of the term informational masking and Carhart’s perceptual masking. Another must read is Spike Tanner’s 1958/64, paper, What is Masking? This paper was required reading at Indiana, Michigan, and Texas during the heyday of TSD in the 1960s (ok, so I’m old), but rarely cited today. This paper is largely about informational masking, but Spike had the good sense to not coin a term for it. Thus, the realization that there is more to masking than just the stimuli has a long history, and if you read Spike’s article you might see that we have reinvented the wheel to some extent.
Back to my statement: I found the work on uncertainty useful and important, and was even more impressed when Donna Neff’s early work with Green on multitone maskers was published in which the role of uncertainty was again shown to have a large effect on performance. I think things began to get off track when people started to make the peripheral vs. central distinction between energetic (peripheral) and informational (central) masking. Chuck was among the first to use this distinction, and I think he was trying to get people to realize that there was more to the auditory system than a basilar membrane and hair cells connected to a nerve bundle. But, what we really need to know are the neural processes that underlie the phenomena associated with masking. Telling me I have to look in the CNS doesn’t help much and certainly doesn’t explain anything.
If things were headed in the wrong direction with the peripheral vs. central distinction, they really went haywire when folks started to use informational masking to describe the decrease in recognition performance of a speech probe interfered with by a speech (or speech-like) stimulus as compared to a noise stimulus (what Carhart called perceptual masking). In this case, it is impossible to be completely sure that one has independent control of energetic and informational masking, since all that is usually done is to change the speech stimulus (e.g., from same gender target and interferer vs. different gender). We went through a terrible period in which informational masking was often defined as not energetic masking which doesn’t explain a thing; or worst yet it was defined as that which doesn’t take place in the periphery (ugh). Durlach and most of the rest of Boston tried to make better sense of all of this in their Letter-to-the-Editor in 2003. But, this only helped me a little. They did go on to start the idea that informational masking comes in two flavors: one related to an uncertainty dimension (a la Watson) and the other to a similarity context (a la speech-on-speech recognition). Chuck’s recent paper (2005) drives home the point that we need to make this distinction. Recently the idea is that informational masking of the similarity kind [IM(S), in Watson’s terms] is tied to sound source segregation. That is, informational masking is a hindrance of source segregation that cannot be attributed to the stimulus per se (especially the peripheral processing of the stimulus). This tends to lead to a chicken and the egg problem, does an inability to segregate cause informational masking or does informational masking cause an inability to segregate? Again, what we want to know is why is the ability to recognize a speech signal (or one of Kidd’s and Mason’s auditory patterns, or other types of “sources”) in the context of a similar source more difficult than when the context is something like a noise. To me this is a really interesting and important question. The answer is not that it is informational masking. Because as of yet, I do not know what similarity informational masking [(IM(S)] is.
OK, so I am unhappy with the current state of affairs. So what. Do I have something different to offer? Glad you asked. Yes. First, let me get my vocabulary clearly stated (a necessity these days if you want to work in this area). I want to distinguish as clearly as I can among masking in a detection task, interference in a discrimination task, and competition in a recognition/identification task. I DO NOT want to call masking, interference, and competition all masking as is becoming common. A detection task is one in which the listener indicates the presence or absence of the signal in the presence of a masker or no masker using any and all aspects of the stimulus. Masking occurs when the detection threshold in the presence of the masker is greater than in the no-masker condition. A discrimination task is one in which the listener must decide which of two targets occurred when the two stimuli differ along a single dimension and/or the experimenter use controls to force the listener to use a single dimension. Interference occurs when the discrimination threshold is greater in the presence of an interfering stimulus than when no interferer is present. In many experimental contexts the task could be described as either a detection or discrimination task. A recognition task asks the listener to use a previously learned label for identifying the probe sound. Performance is measured as percent correct in using the “correct” label (defining correct is an interesting problem, but one for another time). Competition occurs if percent correct recognition is lower when a competing sound is present than when it is not. I would argue that in order to have a greater than chance discrimination threshold the targets to be discriminated must be supra-threshold in the relevant detection task. In order for percent-correct recognition to be greater than chance, the probes must be both supra-threshold in the detection case and supra-threshold in the discrimination case (if both are relevant). Experiments involving informational masking have used all three tasks, but in each case the signal, targets, or probe is always supra-threshold in all ways as defined above. Informational masking is additional masking, interference, or competition caused by one set of maskers, interferers, or competing sounds as compared to another set when the signal, target, or probe are supra-threshold in all regards. There is no need to invoke energetic masking. The signals, targets, or probes are supra-threshold. This is all that is needed. Given my space allocation, I do not have room to expand on this, but let me offer this. There is no or little “informational masking [IM(S)]” if one is asked to detect the presence or absence of a speech probe (or equivalently to discriminate a speech probe from the speech probe plus a competing speech sound). “Informational masking [IM(S)]” occurs when the speech probe is detectable or discriminable (i.e., supra-threshold), but cannot be recognized. Same can be said about IM(U).
Finally, and probably the point I want to have seriously debated: I think all forms of informational masking are a failure of selective attention. We don’t need a term like informational masking (IM) or two terms: IM(U) and IM(S), we already have a useful one, attention. Informational masking occurs when the listener cannot attend to the signal, targets, or probe. Having moved the definition to attention, I can now think of a whole host of things one might try (e.g., cuing) to either aid the subject in attention or further hinder attention. And, I can image what a neural attention network might look like. One example and then I am done. A much overlooked part of Chuck’s presentation on his ten-tone pattern work is the fact that there is no (ok, almost no) “informational masking [IM(U)]” if all that is changed is the task: from same-different to method-of-adjustment. And, once discrimination thresholds in the method-of-adjustment task are lowered almost to the level of performance achieved with no interference, subjects often are able to return to the same-different task and drastically improve their performance relative to what it was previously in the same-different task (we’re talking almost two orders of magnitude in some of Chuck’s conditions). I know of no other way to think about this than attention.
OK-READY, AIM, FIRE
Watson, C.S. (2005). Some Comments on Informational Masking. Acta Acoustica 91, 502-512. Durlach, N.I., Mason, C.R., Kidd, Jr, G., Arbogast, T,L Colburn, H.S., and Shinn-Cunningham, B.G. (2003). Note on informational masking. JASA. 113, 2984-2988. Tanner, W.P., Jr (1958 and 1964). What is masking? JASA 30, 919-921. reprinted and updated as Chapter 24 in J.A. Swets (1964). Signal Detection and Recognition by Human Observers: Contemporary Readings, John Wiley & Sons, New York. I find the chapter in Swets to be more appropriate for thinking about informational masking. Carhart, R., Tillman, W., and Greetis, E.S. (1969). Perceptual masking in multiple sound backgrounds, JASA 45, 694-703. Neff, D.L. and Green, D.M. (19987). Masking produced by spectral uncertainty with multicomponent maskers, P&P 41, 409-415. Kidd, Jr, G., Mason, C.R., and Arbogast, T.L. (2002). Similarity, uncertainty, and masking in the identification of nonspeech auditory patterns, JASA 111, 1367-1376. Pollack’s, I. (1975) paper was an abstract of a paper given at an ASA meeting; not really germane to my discussion.
|