For over a century, our understanding of the nature of taste has been extensively guided by the concept of basic tastes. In the target article, I argued that the use of this model has usually been so implicit that the extent of its influence is not obvious, that the idea itself was not founded on a rational basis, and that it has no testable definition.
But, is my concern realistic for precisely defining the context-giving terms “basic taste,” “across-fiber pattern,” or any other fundamental term, or can we proceed directly to the collection of interesting data without such academic concerns? This turned out to be a large issue for many of the commentators, and thus provided the orientation for many of the responses.
The target article generally emphasizes the necessity of greater clarity in taste research. In analysis of no other sensory system do we find a few sensory “basics”; this is a failed concept in olfaction, and is not the same idea as that used for “basic” colors. I consider that just in raising these issues, the article and the commentaries met their goal. I thank all the commentators for their sometimes surprising, but always thoughtfully illuminating, approaches. I recommend each contribution as worthy of careful study.
R1. Is a theory of taste desirable?
There is wonderment expressed by Di Lorenzo & Chen, Hilbert, and Scott, that I treat research at all levels, from transduction mechanisms to psychophysical organization, with the one model of “basic tastes.” Therefore, a good place to start my responses would be with a justification of this overall monothematic structure of the article.
First, the fact that the taste system functions in a deterministically cooperative way, each level bound with the others, leads to the idea that one model should accommodate the whole. Also, it is clear that for over a century, research in most areas of taste has been guided and constrained by this one “basic taste” model. The experiments from receptor events to behavior have been designed in terms of basic tastes, and the data in all areas came out in terms of the basic tastes.
Whether or not the basic taste term was used, it is clear that that is the basis of how our field evolved and continues. Although rather implicitly, basic taste receptors are fit into the labeled-line model, which has been presented as the neural equivalent of psychophysically defined basic tastes. And this model has the great advantage of simplifying experimental design and thus facilitating the rapid collection of data.
But I believe we will eventually benefit from a properly founded explicit and broadly applicable theory of taste, such as the across-fiber pattern model; something that can give a necessary and testable structure to this field. This model is not designed to pursue particular topics, such as the search for typologies, or the neural representation of specific information, but as an approach to a theory that will encompass all aspects of neural information, including taste. The across-fiber pattern model has been described in detail (Erickson Reference Erickson and Zotterman1963; Reference Erickson1968; Reference Erickson, Schmitt and Worden1974; Reference Erickson and Masterton1978; 1982a; 1984b; Reference Erickson2000; Reference Erickson2001). Its intended role as a guide towards and structure for empirical studies is stated clearly in the prefatory quote from Poincaré.
R2. The basic tastes model reigns
The idea that basic tastes have been the guiding force in the field of taste was noted by Bartoshuk 20 years ago (1988). For example, although in 1916 Henning adamantly opposed the idea of four tastes, espousing instead the idea of a taste continuum (see sects. 4.1.3 and 6.3 of the target article), he illustrated it by labeling the four corners of a tetrahedral continuum with the basic tastes! But with what other tastes or stimuli could Henning have illustrated his model? This may be why Henning is incorrectly quoted as supporting the basic tastes model (Bartoshuk Reference Bartoshuk, Atkinson, Herrnstein, Lindzey and Luce1988), and why research has continued to follow that format.
Does the basic tastes model still give the field its format? Stillman notes we are “hide-bound” by the basic taste orientation, reflecting almost all the commentaries. The only areas of research not affected by this idea might be some applied studies, such as of variations of the threshold for a given basic taste with age, and of the anatomy of the taste pathways; but even these often draw in the idea of basic tastes.
R3. Relating levels
Could we study one area, say psychophysical events, and ignore or even deny the fact that these events derive from the characteristics of the receptors and other factors such as our culture and language (sect. 7.6)? Although a psychophysical study can be successfully conducted without addressing the transmitters involved, that study must not avoid the other known aspects of the taste system. This would be analogous to studying the behavior of oxygen as a reactive gas, as if this were separate from the study of its atomic structure. The usefulness of relating different levels of study is clear in several commentaries.
Di Lorenzo & Chen, Gallo, Kennedy & Gonzalez, Lemon, and Scott usefully relate psychophysical and neural events, as does Hilbert for color vision. Di Lorenzo & Chen intertwine three levels of organization in suggesting that our language may reveal the relationship between neural organization and psychophysics. Belpaeme proposes a direct dependency of language on neural events. Booth conflates many levels of influence into the understanding of taste processing, including, at least, cognitive processes, learning, other senses, and biosocial influences. Sander usefully makes the conflation between behavior (emotions) and neural structures, as well as between taste and the emotions. Whenever I use the term “labeled-line,” I refer to its conflation with “basic tastes.”
Warnings about the problems inherent in drawing casual relations are sounded by Di Lorenzo & Chen, Hilbert, and Scott. Stillman makes the cautionary point that the lack of perceptual basic taste groupings does not negate the possibility of receptor categories. And Cutting carefully refuses to equate the various definitions of basic tastes that do not properly fit together (sect. R5.2).
Clearly, drawing relations between the various aspects of a field is essential for a strong science of taste. But the implicit assumption that the basic taste model justifies these conflations is at best very problematic; this is especially so in its lack of clear linking hypotheses to relate the different levels (sect. 5.4). For example, what links the activity in labeled-lines to the perception of only a few basic tastes? The across-fiber pattern theory explicitly hypothesizes the nature of these relationships (see sects. R10.6 and R10.7).
R4. The role of hypotheses
My view is that good hypotheses are needed for the testable definitions and predictions required in any field of science. Scott's commentary is particularly useful in that our different views of science highlight and clarify each other.
R4.1. On hypotheses and definitions
Scott avers that my concern with detailed hypotheses and “too stringent definitions” of terms get in the way of data collection – which contains the real discoveries. But I doubt that even a simple data collection can be devised without some notion of the meaningful context for the data, and it bodes ill if this context is not spelled out. For example, what does it mean that we have data supporting the idea of neuron groupings, or a cortical area for “sweet?” Are these just data, or does it go deeper than that? The prior, but unasked, questions should have been: “Why should we search for groupings of neurons?” and “What do we mean by ‘group?’” As affirmed in many of the commentaries (sect. R5), we have a poverty of clear definitions, and the underlying hypotheses are very vague at best. Simple data collection is easy, but not simple.
R4.2. On breadth
Again, in distinction to Scott, I believe that breadth is important for a hypothesis. He states that I over-reach in applying the across-fiber pattern theory to other sensory systems, and indeed I apply it beyond input to central mechanisms, as well as to behavior. Several of the commentators also take broad views, including at least Belpaeme, Cutting, Di Lorenzo & Chen, Gallo, Handel, Majid & Levinson, Sander, Stillman, and Roessner,Rothenberger,& Duchamp-Viret [Roessner et al.] (sect. R7; sect. 2.1).
R4.3. On predictions
As suggested by Hilbert and Sander, it requires clear hypotheses to provide the essential predictions to validate an idea (sects. 6 and 7). I welcome these comments. The across-fiber pattern model has provided predictions of core issues in the field that are testable, and these have been met with some success.
R4.4. On control experiments
Although it was stressed as a major problem throughout the article, and was the point of the experiment (sect. 7), there was little commentary on our general and long-term poverty of the control experiments essential for validation of any hypothesis or conclusion. Because disagreement is the best instigator to commentary, perhaps there is no disagreement that, in simply accepting the basic tastes model without test, we do indeed have a long-term and continuing lack of important control studies.
R5. The role of definitions
Let us define our terms. I may have lost most of my audience with that line! I sense that there is little interest in my esoteric enthusiasm for clear hypotheses and shades of meaning of terms like “basic tastes.” Why not just collect the interesting data (sect. R4)? I feel that our routine laxity about the meaning of the core words we use, and the models that employ these words, has produced great problems with much of what we have learned about taste. This view is verified in the differences among the definitions used in the commentaries summarized next. A route towards clarification is contained in the Summary Comment.
R5.1. Are clear definitions desirable?
The first question concerning definitions is whether they are to be desired at all. I was unaware that it is generally recognized that my definitions of basic tastes are too stringent (the same as those of Bartoshuk [1988] and Halpern [2002a]) (sects. 3.2.1 and 5.3.2), as claimed by Scott. He avers that data are equivalent to discovery, that clear definitions slow the production of data, and thus that the accepted lack of a clear definition of basic tastes has encouraged a rapid accumulation of discovery. In this sense he touts the practicality of vague definitions. I cannot agree.
My view that good definitions of terms and hypotheses are of importance is evidently shared by several of the commentators; these include at least Di Lorenzo & Chen, Fox, Gallo, Handel, Hilbert, Kennedy & Gonzales, and Stillman. I have attempted to make the across-fiber pattern model as clear as I can, thus making it possible to suggest various falsifiable tests of it (sects. 6 and 7). Tests of the basic taste model, attempted in section 7, are technically impossible because of its vague definitions.
R5.2. Basic tastes
A central point of this article is that the core idea of basic tastes has wandering and weak definitions. This is confirmed in the variety of ways it is treated in the commentaries.
Scott points out that there is a general acceptance of the fact that we have no usefully clear definitions of basic tastes. In my view, if this is true and we cannot do any better, the concept should be dropped – science questions definitions rather than searching for their support. Kennedy & Gonzalez confirm this vagueness of basic tastes and would like improvement. Booth decries the lack of a good definition and describes a novel and interesting direction that might provide clarity. The previous and present attempts to test the basic tastes idea (sects. 6 and 7) showed that its vagueness prevents a proper test. Scott regrets this vagueness in that it inhibits support of the basic taste idea.
Hilbert details two definitions of basics from color vision that are similar to those used in taste: (1) many colors or tastes can be rated on their basics, and (2) they can be perceived separately in mixtures. He strongly points out that it is not usually clear which definition is implied in taste, including in this article. Logue also relates the idea of basics in taste and color vision. She points out that both tastes and colors can get lost in mixtures; but because there are basic colors, should we not agree on basic tastes? However, Cutting points out, these “basic” terms may not be equivalent. Color basics can be mixed to match other colors, but a failure of this fact in taste may simply mean that these basic terms are not defined well enough for a direct comparison.
Cutting presents a review of the meanings of the “basic” and “primary” concepts; this is certainly useful when considering the meaning of the term “basic taste,” which has also been “primary taste.” He states that a cultural definition of basic tastes makes sense in that it clearly aligns with the needs and desires of the people in their own culture – salt and sugars are important in our lives; this is in close agreement with my discussion of the role of culture and language in our classifications (sect. 2). He suggests that two other definitions have no support at this time: taste mixture studies, as in defining basic colors, and labeled lines. Perhaps he would agree that, if clarified, these two might fit in with the cultural definition. But because of the present lack of clarity, he disagrees with some others who hold that acrid, fat, metallic, umami, and water are basic tastes; indeed, how can we explore for new basic tastes when we do not know what they are?
Sander provides a useful perspective on basic tastes from what turns out to be the closely allied definitions of emotions. Not surprisingly, the number of basic emotions that have been proposed is around six (sect. 2.3), and this position has received support in its analogy with the basic tastes. This analogy makes the basis for the basic emotions especially interesting reading. In a stance similar to Henning's (Reference Henning1916), Sander concludes that emotions are not distinctly different, but like tastes are verbal descriptions of rather factitious points along a continuum.
Warren brings into focus a surprisingly testable but largely untested prediction from the core definition of basic tastes; that is, they may be combined to form all other tastes. He provides an interesting example of how this prediction can be tested. But the number of successes can be counted on one hand. I would guess that there have been very many tests of this most obvious and important prediction from the basic tastes model, but that those scientists found it did not work, and did not publish. But falsified predictions are at the heart of science, and these would have been most prominently published were this physics. We do not have a Journal of Negative Results, but imagine how informative one could be! Perhaps an internet JNR?
R5.3. Groups/types
The idea of basic tastes strongly leads to the assumption of underlying anatomical basic taste groupings or types, such as labeled lines or specific receptors (sect. R3). My considerable concern with groupings centers on the ease with which groupings of many sorts can be developed absent a critical rationale (sect. 3.2).
As examples, Fox and Gallo accept groupings of specific receptors, and they and Di Lorenzo & Chen support labeled-line groups. Scott has searched for neural groups with mathematical analyses such as multidimensional scaling and cluster analysis, and then when those failed (sect. 4.2.2), he searched for other support and found it in an influence of motivational state on neural responsiveness; Fox suggests that the latter definition supports the former. These commentators also cite successful anatomical and receptor classifications. But an inspection of a clearly nonbasic continuous sense, audition, shows that if only four disparate tones were used, say 500, 1,000, 1,500, and 2,000 Hz tones, they serve well as “best stimuli” for auditory neurons, and a different neural cortical location would be found for each; then are these “basic” tones? There is probably a neural area, best-neuron type, and receptor type for lysine; but this control was never performed because lysine is not a “basic.” We find what we look for.
Fox claims that the across-fiber pattern model fails in that there is no sufficient definition of the groups necessary for it. He does not note that groupings define the labeled-line view, whereas in the across-fiber pattern model they are only necessary in nontopographic systems such as color vision and taste, where they have been defined.
Lemon and Lavine offer strong caveats against accepting mathematical analyses as support for the idea of groupings in taste. Lemon points out that the measures of neural responses as seen in repeated measures, even in the same animal, are inherently too unstable to properly define “best stimuli,” and comparisons across animals result in additional problems. And, of course, best stimuli change with variations in stimulus intensity, and with the stimuli chosen for searching for the best stimulus; if only basics are used, one of them must be best. Also, Lavine makes the point that the common use of certain mathematical procedures to support the classification of data into groups is invalid; he cites multidimensional scaling, cluster analysis, and factor analysis as culprits. Fox agrees, as do I (sect. R6; sect. 4.2.2).
It is probable that the search for support of classifications has been driven by desires to demonstrate that basic-tastes groups exist. A motivated search for groupings, without strongly specifying what a relevant grouping would be, is bound to succeed.
R5.4. Labeled lines
The following definition of “labeled line” generally follows that suggested by Di Lorenzo & Chen, Fox, Lemon, Scott, and Stillman: A labeled line is a neuron whose activity is labeled in its meaning. In taste the meaning is to represent one of the basic tastes. As I understand it, the meaning of activity in a labeled line is set whatever stimulus evoked the activity, be it the best stimulus or other, and it is not interpreted in terms of activity in other neurons (as in the across-fiber pattern model), or by the temporal pattern of its activity. The label is fixed. But perhaps this definition could use some serious tuning-up.
In support for the labeled-line idea, Di Lorenzo & Chen, Fox, and Stillman suggest that a broadly tuned neuron's “best stimulus” could make it useful as a labeled line. Di Lorenzo adds that even if there are many more tastes than just the basics, there could be a labeled line for each. Depending on the number of separate tastes, this could encumber many labeled lines in a relatively sparse sensory system. Her quantitative prediction is worth study. Scott accepts any relatively narrow tuning as a labeled line.
In comparison with the across-fiber pattern model, these definitions absolutely distinguish the labeled-line from the across-fiber pattern model in that in the latter, each broadly-tuned neuron participates with others in the representation of many different stimuli, and responses to all stimuli, of whatever magnitude, are accepted as information. The identity of the best stimulus is not relevant as in the labeled-line view. I discuss labeled lines in the article as the extension of an across-fiber pattern model to a homogeneous group of neurons wherein the information is reduced to one bit – only one message (sect. 8.4.2).
In criticism of the labeled-line idea, Lemon notes that “best stimuli” are too unreliable to serve as indicators for labeled lines. Then what can the meaning be of activity in a labeled line? And if a broadly tuned neuron's meaning is considered to be only that of its best stimulus, responses evoked by other-than-best stimuli would produce confusion and wasted effort for the nervous system. On the other hand, broad tuning is accepted “as-is,” an integral and functional part of across-fiber pattern coding.
Kennedy & Gonzales quote one of the wise old men in taste, Dethier (1974), that the “rigid specificity” of labeled lines “existed more in the minds of investigators than in the receptors themselves.” This rings a bell with Sander's comment that the six basic emotions perspective is “an article of faith rather than an empirically or theoretically defensible basis.”
Beyond that, Scott and Stillman state that few scientists are concerned with a good definition of labeled lines. They do not give a reference, but if this is true, is there no real, good, and generally accepted definition of labeled lines? Does no one care? I suggest that a critical and testable definition should be developed before we are led further by this idea.
R5.5. Across-fiber pattern
The definition of information representation in the across-fiber pattern model is simply this: The information in a neural message is defined by a unique pattern of activity across neurons. Details of this idea are given in sections R9 and R10, and sections 6, 7, and 8 of the target article. Booth makes a perceptive and exact statement of one important aspect of the across-fiber pattern model; that is, the neural patterns hold from receptor to behavior without “read-out” points in their journey (Erickson Reference Erickson2001). It is nice to have that spelled out.
R5.6. Labeled-line versus across-fiber pattern?
Fox and Di Lorenzo & Chen suggest that the two models are largely indistinguishable because they are both spatial, and because of the lack of strong definitions to separate them. This alone should make clear how essential stringent definitions are to our science. Di Lorenzo & Chen also suggest that labeled-line and across-fiber pattern coding are not different in that the responses in a broadly tuned neuron may be interpreted as a labeled line according to that neuron's best stimulus. Fox and Kennedy & Gonzalez make the same point, while opting for both the across-fiber pattern and labeled-line models. Booth reasonably questions both labeled-line and across-fiber pattern models to the extent that there is a lack of good definitions.
Several commentators pose the possibility that labeled-line and across-fiber patterns can exist in parallel. As Kennedy & Gonzalez point out, this idea has been around for a long time; they align with Dethier (mid 1970s) in proposing two coding mechanisms corresponding to (a) acceptance/rejection and (b) the encoding of many taste stimuli. The first might be called two labeled lines providing the minimal information required for simple yes or no responses (sect. 8.4.2). For the second, Dethier invoked the across-fiber pattern code. This sensibly suggests that the two ideas are distinctly different, while possibly coexistent.
R5.7. Taste continuum
Whether or not taste can usefully be defined as a continuum is a clear point of distinction between the basic-taste and across-fiber pattern models; the former cannot tolerate a continuum, whereas the latter embraces it. But, as is clear from the commentaries of Handel and Stillman, it is hard to believe that taste could be a continuum (sect. R9). Sander supports the idea of a taste continuum in its similarity to a continuum of emotions (sect. R7.3), and I find it probable in taste (sects. R5.8, R9, and R10.3; sects. 6.3 and 8.4.2). So this is an important but very tricky issue. Give me a moment.
Because chemicals are discrete entities, how could they be considered as a continuum? The elements find their best organization when considered by mass, as in their representation in the periodic table. They do not form a continuum, but are most productively considered as being members of a continuum. This is how I use it in taste. I think it is premature to assert that there is not a continuum of taste without test.
The presence of a taste continuum is evident in the neural data (Erickson Reference Erickson, Kare and Maller1967). When considered beyond the four basic stimuli, the neural responses are seen to conform to neural response functions tuned broadly across some continuum, much like the color receptors (sect. 6.3). If we pretend that we do not know the wavelength continuum, the responses of color receptors can be used to generate an illustration of that continuum, as well as the bell-shaped characteristics of the color receptors. In this pretend situation, physicists could be asked to define the color continuum. These same methods generate a taste continuum from the neural responses of taste neurons, as well as demonstrate their bell-shaped neural response functions across this continuum. Given a broad enough range of stimuli – certainly beyond the four basics - chemists could be asked to identify this continuum. The fact that mathematical solutions such as multidimensional scaling (sect. 4.2.2) succeed at all in taste – even if in a distorted and not clearly interpretable form (e.g., not being able to prove the existence of neural groups) – indicates that there is indeed a useful underlying continuum. If taste data are viewed in this manner, the chemistry, neural organization, and psychophysics of taste may gain clarity they would not get if we just stick to the basics.
Di Lorenzo & Chen claim that labeled lines are not incompatible with a continuum; they could just appear at certain points along it. But this is hard to rationalize with the definition that activity in a labeled line is unrelated to activity in other labeled lines, whereas a continuum is a specification of relationships.
R5.8. Breadth of tuning
The terms “narrow” (or specific) and “broad” are commonly used to describe the tuning of individual afferents. It is certainly not always clear what those terms mean, or why we should care. “Broad” seems to refer to neurons sensitive to large portions of the stimulus continuum, such as with color receptors, and “narrow” indicates neurons sensitive to a small portion, as in tactile or visual location.
But a more explicit prediction for breadth of tuning in the across-fiber pattern theory is in order here (sect. 6.4). In this model, proper breadth of tuning is a strictly mathematical issue. The tuning should be sufficiently broad to gain enough neural mass to make the necessary discriminations. What is enough mass? In color, the few neurons available at each retinal point must be just broadly tuned enough to provide the mass out of which can be carved the sufficient neural mass differences required for signaling changes in wavelength. On the other hand, in representing visual location, there are so many neurons available that broad tuning across space would result in an unnecessary, indeed overwhelming, incoming neural mass.
In finer detail, quantification in the across-fiber pattern model suggests that any discriminable change, wavelength, location, taste, or other, depends on the same amount of neural mass difference – a constant, X. In the uniform darkness of the brain, where one system cannot be seen as different from another, just X is what the nervous system needs to detect a change, whether visual location or taste. It can just notice X. This quantifiable hypothesis can be tested.
Back to the vernacular, that taste receptors and neurons are broadly tuned seems to be accepted by almost all but the molecular biologists (sect. 5.3.1), and now the molecular findings have been reconciled with this clearly established breadth (Tomchik et al. Reference Tomchik, Berg, Kim, Chaudhari and Roper2007). But the role of this breadth is debated. None of the commentators deny the necessity of broad tuning in the across-fiber pattern model for taste, and Logue points out that such breadth is contrary to the labeled-line code. But Di Lorenzo & Chen, Fox, Gallo, and Kennedy & Gonzales suggest that better sense could be made out of broadly tuned neurons if they were considered as labeled lines according to their best-of-the-basics stimuli (sect. R10.2).
R5.9. Singularity
The labeled-line model indicates that only individual basic tastes should be perceived as singular, whereas the across-fiber pattern idea predicts the singularity of many tastants and mixtures. Tests of this difference should be a primary effort for evaluation of the models, but they are seldom attempted. Therefore, I find Gallo's attempt to test one of these core aspects of basic tastes particularly gratifying. (Warren tested the other core definition (sect. R5.2). More tests should be considered essential, and are not complex.
As context, I have shown that human subjects may sometimes classify an individual stimulus or a mixture as singular, and at other times as more-than-one (sect. R9; sects. 3.2.1, 4.1, 6.1, and 7.9). Using rats, Gallo finds that although complex stimuli can be rated as singular, as in the across-fiber pattern model (sects. 6.1 and 7.9), the basic components of a mixture can also be identified, supporting the basic taste idea. The result depends on the context. To provide him with a possibly passable example, a complex object, such as a car, might be seen as a single object, or as more-than-one (windows, tires, or such), depending on how the question is asked. But if simultaneous activity in four labeled lines were reported as singular, that would be equivalent to perceiving them as an across-fiber pattern. This is a topic in taste that could use further study.
R6. Misleading methodology
The basic taste methods have uncritically supported the basic tastes model, and thus have certainly distorted the field. To illustrate this position, imagine a color mixture experiment in which the subjects are limited to responding only with the basic color terms, red, green, yellow, and blue. When asked to rate a mixture of green and blue, they are only allowed to respond with the words green and blue, and cannot say “a singular hue something like blue and green, but not exactly either.” If allowed, they would also respond that these colors are reduced in basics saturation, as less blue or less green; this is illustrated in studies of color-naming (Erickson Reference Erickson, Le Magnen and MacLeod1977). This procedure is standard in taste studies; the subjects are not allowed to rate a taste mixture of HCl and NaCl as a unique taste, but must respond in terms of the basic tastes, sour and salty – thus always supporting the basic taste idea.
This pervasive but faulty methodology of using only basic tastes has led to strongly biased support for the basic tastes model. This one surprising error appears to have largely provided the shape of the field as we know it.
As a good example of this bias, Kennedy & Gonzales point to the fact that the molecular studies (sect. 5.3.1) supporting the idea of specific basic taste receptors would very probably have shown broad tuning if a range of stimuli beyond the basics had been used. They state that in studies of the receptor molecules, the researchers “assumed the hypothesis of basic tastes and collected data to support it and the labeled-line hypothesis, but did not test the hypotheses.” Majid & Levinson also cite the need for a larger array of taste stimuli beyond the basics. I suggest the employment of systematic variations of taste stimuli of some variety, at least beyond the four basics (sect. 5.4). And now it has been shown that the receptor cells are broadly tuned (Tomchik et al. Reference Tomchik, Berg, Kim, Chaudhari and Roper2007).
There have been many attempts to find the groupings required by the basic taste model. Many researchers, including Scott herein, have sought these groups through improper use of mathematical techniques such as multidimensional scaling, cluster analysis, and factor analysis (sect. R5.3; sect. 4.2.2). Fox, Logue, and Scott cite the findings of CNS areas devoted to particular basic tastes as supportive of the basic tastes model. It is probable that when searching for labeled areas, using basic tastes or others, they will be found. The necessary controls using other than basic tastes are lacking.
All such findings supporting the basic tastes idea, and most of those of the last century, were directed by the context of the basic tastes hypothesis in which they were generated. Thus, this model has very extensively and uncritically supported itself.
R7. Informing ourselves more broadly
The value of a broad view of taste is emphasized in several commentaries, such as those by Booth and Gallo, and others mentioned next. They represent the systems approach to science in distinction to the reductionistic or molecular approach. Just as psychophysical, neural, and receptor events must inform each other, related areas of study also provide important insights into the realities of the nature of taste. However, Handel cautions that comparisons between systems can be problematic when the systems differ, and Scott laments the broad applicability of the AFP theory beyond taste (sects. R4.2 and R9).
R7.1. Insights from studies of language
A primary thesis of the article is that the discrete nature of our language may inappropriately drive us to use the concept of a few discrete basic tastes (sects. 3.2.1 and 4.1). But several commentators also point out that language may properly elucidate certain aspects of taste.
Di Lorenzo & Chen usefully intertwine our words, psychophysics, and neural organization, such that the basic taste words may reveal our perceptions of genuine aspects of neural organization. This idea would be interesting to formalize as a testable hypothesis. Their position relates to the relativism and nominalism of Belpaeme and Majid & Levinson, discussed next.
From studies of color-naming, Belpaeme (sects. R3 and R9) takes a linguistic-relativism position that our language modifies innate biologically driven perceptual organization, and can shed light on this organization. He contrasts this with my universalistic approach, which is that the taste sensations are driven only by biology. He describes procedures in color-naming that could be useful in taste studies.
In a related view, Majid & Levinson's nominalist position is that words actually create valid basic taste categories (sect. R5.2), and that this has cross-cultural support. For example, the same confusions between bitter and sour exist across cultures, and common patterns of taste words across cultures support the idea of umami and others as basics. He contrasts this with my realist position, which is that events – here tastes – exist independently of our naming them. These views of Majid & Levinson and Belpaeme seem strongly related to that of Cutting (sect. R5.2) on the force of our culture on the organization of taste.
This commonality of taste perceptions across cultures is neurally meaningful. For example, bitter and sour stimuli produce somewhat similar across-fiber patterns, and thus they can be confused with each other. Why they have separate names is important, but that does not mean that they are as discriminable from each other as either is from a sweet stimulus. Useful taste names should not be taken as a rejection of a taste continuum any more than color names are a rejection of a color continuum; but neither should these names be taken as proof of only a few basic sensations.
On the other hand, in his unique search for definitions based on his experience with color vision (sect. R5.2), Cutting shows how only the cultural definition provided by language gives support for the idea of basic tastes, and that the definitions based on stimulus mixtures and labeled lines are not supportable. He raises the good questions of how one definition may be compared to others, or support be found between them, and comments that they certainly should not be confused with each other.
R7.2. Insights from the study of emotion
Sander describes several useful and interesting parallels between the studies of emotion and taste. In a comparison with taste, there are currently considered by some to be six basic emotions; this is directly in line with Miller's prediction of about that number (sect. 2.3). The idea of neural areas for basic tastes (sect. R5.3) has suggested that neural areas may also be found for specific emotions. Sander shows these arguments to be problematic, and that a continuum of emotions is a more reasonable view, as suggested for taste in the target article (sects. 6.3 and 8.4.2). As the two fields seem to be somewhat in the same position, he suggests that better communication between them would be helpful.
Importantly, the prior categorization of emotions into basic groups is being questioned rather than assumed, an obviously good tactic for basic tastes.
R7.3. Insights from a broader view of sensory input
Many commentators advocate the advantages of taking broader views across sensory systems. Handel makes useful comparisons with audition, and Belpaeme, Cutting, Fox, Handel, Hilbert, Logue, and Stillman with color vision. The AFP theory applies across sensory systems (sect. 8.4.2).
As an informative example of the utility of a broad point of view, Roessner et al. point to the relationships between taste and its close ally, olfaction. We have much to gain from careful comparisons of the two. Obviously, taste stimuli have olfactory components, taste and olfaction cooperate in food intake and other functions, and they are both chemical senses.
Given this close cooperation, it is informative to note that they are anatomically very separate at the input end; taste is directed into the hindbrain and moves anteriorly into the forebrain, in company with vision, audition, and somesthesis. Olfaction, the most ancient of the senses, enters and forms the most anterior part of the brain. But even though separate, taste and olfaction seem driven to cooperate; how and why this cooperation happens is certainly worthy of study.
Olfaction is not the only sense that cooperates with taste. For example, it is clear that input into the hindbrain taste relay comes not only from olfactory input (Van Buskirk & Erickson Reference Van Buskirk and Erickson1977a), but also from the nasal trigeminal system (Van Buskirk & Erickson Reference Van Buskirk and Erickson1977b) and from the stomach (Glenn & Erickson Reference Glenn and Erickson1976; sect. 8.5.4). The study of taste would certainly benefit from a broader view of interactive sensory inputs.
R8. Critique of the basic tastes model
Some critiques of the basic taste format are brought together here in brief form. Foremost, this model has directed research and understanding throughout the history of taste. The simplicity and convenience of this model has made it very attractive. However, Stillman suggests that the labeled-line idea may not be as broadly accepted as I indicate.
As a primary criticism of this model, Kennedy & Gonzalez note that good definitions are an essential requirement for the evaluation of models, and that the basic tastes model is quite undefined. The lack of defined linking hypotheses for the interactions between different areas of research is part of this problem (sect. 5.4). But as Scott notes, these weak or missing definitions facilitate support for the model.
The lines of evidence that have been used to support the basic tastes model include all psychophysical studies that use only basic tastes (sect. 3.2.1); the idea of groupings (sect. R5.3; sect. 3.2.3); the labeling of broadly tuned neurons or neural areas by their best stimuli by Di Lorenzo & Chen, Fox, Logue, and Scott; and Logue's conflation of basic colors and tastes. Cutting finds support for basic tastes only in their cultural definition.
R9. Critique of the across-fiber pattern model
As with the critique of the basic tastes model (sect. R8), the relevant aspects of the AFP model that were detailed elsewhere are briefly compiled here.
Concerning the breadth of the AFP model's applications, Scott (sect. R4) complains that I over-reach in proposing that all sensory messages are encoded in patterns. But a good hypothesis covers large areas of investigation, thus providing parsimony of concepts. In this sense, the across-fiber pattern is a good model (Erickson Reference Erickson2000; Reference Erickson2001) in that the nervous system will probably stick with a good method of representing information once it has found one, and the AFP model applies to all systems; both vision and taste, as well as movement, are built on the same principles, as are fish, fowl, and elephants. Of course, retinal disparity and sensitivity to differences in timing of auditory inputs are quite unique, but still the AFP model is appropriate for them.
Handel, Hilbert, and Stillman doubt that the across-fiber pattern model is appropriate for color vision at the central levels (beyond the receptors); however, these neurons are as broadly tuned across wavelengths as are the receptors, and thus fit the AFP model. Sander's review of the study of emotions suggests the across-fiber pattern model's breadth of application.
Lemon raises the issue of quantifiability as important for a model. The AFP model is clearly quantifiable, giving predictions for test (sects. 6 and 7).
Young's model is based on the idea of economy of neural resources, so this very important asset is part of the AFP model (sect. 8).
Predictions are an essential part of a good hypothesis (sect. R4; sects. 6 and 7). The AFP model is strongly predictive of a variety of findings. These include the probability of a continuum underlying the sense of taste (sect. 6.3), denied by Handel and Stillman, but supported by Sander via the emotions; broad tuning of receptors and neurons in some nontopographic systems such as taste; narrow tuning of receptors and neurons in topographic systems such as location in somesthesis and vision; compatibility with a temporal code; and the existence of types of receptors and afferents in nontopographic modalities such as color and taste. Scott has over time cited the evidence both for and against typologies as denials of the across-fiber pattern model (sect. 5.4).
The idea that taste mixtures could be perceived as singular has been successfully tested (sect. R5.9; sects. 6 and 7). And Hilbert points out that colors are singular – a successful prediction from the across-fiber pattern theory.
R10. Misunderstandings about the across-fiber pattern model
The AFP theory is very largely misunderstood. This is clearly borne out in the diversity of viewpoints expressed in the commentaries. I hope that this brief discussion will provide some clarity.
R10.1. Typologies/groupings
A number of important differences appeared among the commentaries on the relationships between across-fiber pattern coding and groupings.
The AFP model is designed for the representation of information broadly throughout the nervous system, not just for taste (sect. R58; sect. 8). The general requirement for typologies in this model derives from the paucity of receptors or neurons available in nontopographic modalities, such as color vision and taste. Also, the body might find it too genetically expensive to generate many different pigments, each specifically or maximally sensitive to only a very small section of the continuum – or to generate a receptor for each tasteable listing in the Merck Index. Thus, the representations of colors and tastes are argued to employ a few broad and cooperative types of receptors and neurons.
On the other hand, in the topographic modalities, there are very many neurons available to encode the spatial maps of location across the retina and skin. It would be anatomically difficult to have each of these be broadly tuned across the body; for example, a “nose-tip-best” neuron extending its gradually descending sensitivity down to the ankles. And, because of the generosity of neurons, there would be sufficient neural mass generated by point stimulation of the skin to provide for discrimination of location (sect. 6.4). For both reasons, groupings are not called for, and are indeed counter-productive, for topographic modalities.
Thus, as pointed out from 40 years ago (Erickson Reference Erickson and Zotterman1963; Reference Erickson1968) to the present (Erickson Reference Erickson2000; Reference Erickson2001; and the present article), the AFP theory is not silent on the issue of typologies as Scott avers, but instead is very explicit. The intent is to be sufficiently explicit that the idea can be further considered and tested.
Di Lorenzo & Chen point out that although groupings have not been defined or demonstrated (Lavine), it does not mean that they will not be; certainly true (sect. 8.5.1). But instead of a search for any typology, which must succeed, I hope that the rationale for groupings and their definitions will first be sufficiently explicated that the idea can be tested. Groupings are not contrary to across-fiber patterning and are the definition of labeled-line coding (sects. 4.2 and 8.5.1).
Lemon suggests that the AFP theory needs groupings to get unique patterns of neural activity. But unique patterns underlie all discriminable inputs whether or not there are groupings. The across-fiber patterns for each discriminable tactile stimulus or auditory frequency are unique, but there are no groupings of these neurons.
Interestingly, Handel gives a rationale different from Young's for the neural typologies in color, and the lack thereof for auditory tones. He points out that differences in the physical characteristics of the stimuli demand three types of color receptors and many types for tones. The fact that the former is nontopographic and the latter is topographic leads to the same conclusions from the AFP theory of taste. How these two disparate orientations come to the same conclusion is an interesting question.
R10.2. Breadth of tuning
The fact that neurons are broadly tuned is generally accepted, but it is often a cause of broad concern. It may be that it is intuitively difficult to see how this breadth is anything but noise that needs to be silenced.
Di Lorenzo & Chen take this view in their claim that breadth of tuning results in a loss of information. This is addressed in section 6.4 where breadth is shown to be the basis for the subtle coding of large amounts of information, and previously, where it was shown that breadth of tuning causes no loss of information (Erickson Reference Erickson1968, Fig. 2). In brief, as breadth increases, the neural mass increases, whereas the neural mass differences – the basis of discrimination – remain constant (sect. 6.4). This happens because recruitment of additional neurons as breadth of tuning increases just compensates for the loss of neural mass differences obtained from each neuron.
One approach to clearing up the noise, as voiced by Di Lorenzo & Chen and Fox, is that the nervous system might interpret all activity in broadly tuned neurons – whatever stimulus caused it – as representing their “best stimulus,” making them labeled lines. Similarly, Stillman defines any well-identified group as representing labeled lines, even if broadly – tuned, such as color receptors. But even with such blunt tuning, the color receptors still provide for acuity approaching 1 mµ – expected only in the AFP model.
Scott accepts any narrowly tuned neuron as a labeled line. He claims that I should admit that this narrowness of tuning, evident in topographic systems, is an exception to across-fiber patterning. But the tuning there is still much too broad to account for the accuracy of tactile localization (sects. R5.8 and R10.1), as was noted by Adrian, Mountcastle, and Sperry, as well as many others (see Erickson Reference Erickson2001); each of these scientists saw the need for a version of across-fiber patterning even for this “narrowly tuned” system.
Both broad and narrow tuning require AFP coding (sects. 8.4 and 8.5).
R10.3. Color vision
The analogy between the neural coding for color and that for taste is a fairly common theme among the commentators. That the form of the neural representation of color changes from three bell-shaped neural response functions at the receptor level, to a two (or four) opponent process representation beyond, causes Handel, Hilbert, and Stillman to question whether the AFP model could hold throughout the visual system. The tuning is broadly bell-shaped in the receptors, and broadly “S”-shaped centrally, with the Ss lying down across the wavelength dimension, inhibitory towards one end of the wavelength continuum, and excitatory towards the other (De Valois Reference De Valois1960).
The important aspect of neural response functions for color is that they be broad and simple, and indeed this is their characteristic peripherally and centrally. The AFP model is not concerned with the shape of the function, and thus is equivalently competent to represent wavelength both at the receptor level and beyond. Color is not the only dimension represented by other than simple bell-shaped neural response functions; for a review see Erickson (Reference Erickson2001).
R10.4. Temporal codes
Di Lorenzo & Chen criticize the AFP model for excluding the possibility of temporal coding, and Lemon also raises this question. However, the across-fiber pattern model includes temporal coding (sect. 8.5.2; Erickson Reference Erickson2001). On the other hand, and from a more neutral perspective, it might have been pointed out that labeled lines cannot use information in a temporal pattern because their meaning is defined only by their best stimulus (sect. R5.4). The equation of both labeled lines and across-fiber patterns as spatial models (sect. R5.6) glosses over their very substantial differences, and should not be used to rule out temporal patterns in across-fiber pattern coding.
R10.5. Quantification
An important asset of the across-fiber pattern model is that it is inherently quantifiable, for example, in providing predictions about intensity thresholds, discrimination between intensity levels, and discriminability among different tastants (sect. 6.4). But Lemon suggests that some stimuli present rather similar across-fiber patterns that are not compatible with their perceptual distinctness. For example, although HCl and QHCl can be easily discriminated, the across-fiber patterns they produce are not as distinct as that caused by many other pairs of stimuli. There may be two issues here. On the one hand, these stimuli are similar in eliciting avoidance, and thus some similarity in their across-fiber patterns might be expected. On the other hand, they are motivationally charged stimuli, encouraging a high level of behavioral differentiation between them when needed. Gallo's data on the rat's ability to pick out individual taste cues in a mixture suggests that quinine and HCl might be highly discriminated when this is demanded. Whatever the case, the across-fiber pattern model lays itself open to tests of its quantitative estimates of the differences between stimuli.
R11. Questions and comments about the experiment
This article is intended to advocate for hypotheses and basic terms that are sufficiently clear that they can be properly tested. Such a control study is offered to make the possibility of tests clear (sect. 7). This preliminary study is intended more to encourage other more fulsome studies than to settle the questions asked.
Several commentaries on this study suggest that there is a strong tendency to accept the basic taste idea rather than to test it. For example Booth, and Di Lorenzo & Chen, as well as others who previously viewed the article, point out that I should not have claimed that the nonbasic tastes could account for the various comparison stimuli; instead, I should have acknowledged that these tastes are effective only because they are composed of the basic tastes. This criticism assumes the validity of the basic tastes idea while testing it! It suggests that control experiments are not called for because we know the basic tastes idea is true! But, by the neutral position required in this situation, it could as well be claimed that the basic stimuli are only effective because they are composed of the innately effective nonbasic stimuli (sect. 7.10). This demonstrates that the basic tastes idea does, in fact, exert strong control over our research and understanding of taste. It is so strong, that there is evidently no desire to perform the required control tests.
Di Lorenzo & Chen state that, because the accounting by the basic tastes was not total, I infer other tastes. I do not mean to suggest this. Instead, I suggest that there may be a lack of clarity in tastes that limits complete accountings (sect. 7.7.2). And I suggest that a “completion effect” may interfere with the strength of any accounting seen (sect. 7.8, and the Appendix).
Di Lorenzo & Chen also cite the common view that the variety of words offered by the subjects in the experiment to describe the tastants (sect. 7.7.3) indicates that nongustatory inputs (thermal, tactile, etc.) give these tastants their distinguishing character; they are simply composites of the basics adulterated by nontaste inputs. It may be that this defense of the basic position is used more often than the missing facts might indicate, and should be examined in each case rather than assumed.
Logue suggests that the basic tastes idea is supported by the data because the basic taste stimuli (but not the basic words) are better than the nonbasic stimuli. But the definition of basic tastes (sect. R5.2) is that they, and only they, totally account for other stimuli. Then the model fails because (a) the accounting by the basics is not total; in fact, the basic words do rather poorly, and (b) the nonbasic stimuli are rather effective – they certainly do not lie along the baseline as good nonbasics should. Also, she points out a fault that the data were not treated statistically; but the definition of basics is absolute rather than statistical, as noted by Hilbert; what statistical tests would be appropriate?
Hilbert correctly points out that the protocol using words has no real control. However, the use of the basic words and stimuli as equivalent agents of the basic taste model permeates the literature, so they might well be compared as in this experiment. Still, I agree with his complaint, and hope the efficiency of a variety of words and stimuli to account for different tastes receives further study in unbiased, controlled settings.
R12. The “across-fiber pattern” term
Lemon uses the term “across-neuron pattern” and Warren uses “cross-fiber pattern,” both evidently referring to the same idea as in “across-fiber pattern” – but I am not at all certain about this. And a number of other terms outside the field of taste are being used that probably have the same meaning as “across-fiber pattern,” such as “combinatorial” coding in olfaction, and others (sect. 8.4.2). I suggest that to the extent that several terms refer to the same idea, it would be helpful to use the same term. This would also have the beneficial effect of requiring stringent definitions of each of these terms to determine if they do indeed refer to the same idea.
I would like to clarify my role in the across-fiber pattern model. Logue attributes the term “across-fiber pattern” to Pfaffmann (Reference Pfaffmann1941). Although he discussed the implications of patterns of activity in parallel neurons (1959), Pfaffmann did not claim that term. Throughout his career he was primarily concerned with the meaning of activity in broadly tuned “best stimulus” neuron types, which he considered to be labeled lines (Pfaffmann Reference Pfaffmann, Carterette and Friedman1978; Pfaffmann et al. Reference Pfaffmann, Frank, Bartoshuk, Snell, Sprague and Epstein1976; Reference Pfaffmann, Frank and Norgren1979). This idea was developed further by his student, Frank. My role was to realize the great strength and broad applicability of the across-fiber pattern idea, to realize that it had been discovered by many scientists in many or all neural systems since Young – and continues to be discovered, and advocated that it be given one name to the extent to which it is indeed one idea (Erickson Reference Erickson and Zotterman1963; Reference Erickson1968; Reference Erickson, Schmitt and Worden1974; Reference Erickson and Masterton1978; 1982a; 1984b; Reference Erickson2001).
R13. Conclusion
The commentaries reinforce my view that we are indeed in the grip of the obscure and overshadowing idea of basic tastes; this gray eminence directs our research and thought whether we are aware of it or not. We clearly do not know what this core concept means, but we are so comfortable with it we do not want to raise any questions. We certainly have not tested it, even though testing of the core hypotheses – as distinct from data collection – is an absolutely essential part of science. But if someone thought to test it, how would they start? They would face the basic problems Hanig (sect. 3.1) and Henning (sect. R2; sects. 4.1.3 and 6.3) faced; what stimuli other than the basics could they use, and against what other hypothesis would they test it? “But what else could we do?” should not be a problem, but a realization of the necessity to search for a clear direction.
Such issues, which I have tried to emphasize for over 40 years, appear of little interest to researchers in taste; for example, although the across-fiber pattern theory is clear, no definitions for “basic tastes” have been forthcoming, or even concern shown. But if anyone is indeed concerned, I have a suggestion. Someone could start an on-line Gustopedia. Someone might offer a testable hypothesis, or a definition, or a statement of how one definition or level might relate to another. Another person could add their constructive comments. Studies from other areas of science could add perspective. The expression of differences would be most useful. Eventually some clarity could evolve to provide the firm basis required of any science.
