Rosch: "Principles of Categorization" (1978)

Elanor Rosch's contribution to Cognition and Categorization really consists of two independent parts: An overview over her experiments investigating basic-level categories (pp. 30-35), and an overview of her experiments with prototype effects (pp. 35-41). In will only deal with the first part now.

Cue Validity

The most central concept in Rosch's discussion of basic-level categories is the notion of the cue validity. This is defined for a category such as "bird," which is more or less reliably identified by cues such as "wings." She explains:

The cue validity of an entire category may be defined as the summation of the cue validities for that category of each of the attributes of the category. (pp. 30-31)

This immediately raises two questions:

1. Do all cues count in the summation with equal weight? There are infinitely many possible cues and only a few highly valid ones. This suggests that more explicit assumptions about "salience" are needed.
2. With what weight do the various members of a category contribute to the average? Equally? Weighted by the frequency of the linguistic label? Weighted by the frequency of the thing?

While these questions may seem like technical remarks, they do in fact relate to some deeper issues that I will mention below.

The Ambiguity of "Basic"

There are two competing characterizations of "basic" in Rosch's work, an ostensive and a perceptual. It's not always clear which one she is taking as definitive, and this sometimes introduces problems.

Both definitions apply to concept trees and are meant to pick out a particular depth in such a tree. They do so by locating the level of abstraction at which either

1. the categories "car," "chair," "tomato," and "hammer" are found; or
2. average cue validity is maximized.

My worry is that her cross-cultural, developmental, and evolutionary claims may turn out to be tautologies when we look closer at the ups and downs of her theory.

For instance, if the "basic" means "maximal cue validity," then of course children learn names from this level first. On the other hand, if Rosch gets to pick what counts as "basic" in each branch of the English category system ("chair," "car," "tomato," ...), then she can obviously just pick the level that fulfills the second definition.

Learned Perception

The fact that she might unknowingly be making the tautological point that "normal things are normal" is hinted at when she comments that English-speakers tend to be less able to distinguish between plants than the ostensive definition suggests.

This is observation was echoed more recently my Jerome Feldman:

For many city dwellers, tree is a basic category—we interact the same way with all trees. But for the professional gardener, tree is definitely a superordinate category (Feldman 2006: 186)

With those kinds of qualifications, basic level categories will certainly guaranteed to have all of the properties that Rosch claims. But any claim about their universal centrality will also become an empty verbalism.

Note how this also ties in with the sticky issue of trained perception:

One influence on how attributes will be defined by humans is clearly the category system already existent in the culture at a given time. This our segmentation of a bird's body such that there is an attribute called "wings" may be influenced not by perceptual factors [...] but also by the fact that at present we already have a cultural and linguistics category called "birds." (p. 29)

She is apparently aware of this problem, but not willing to face the implication that complex cues like plumage are themselves categories that are open-ended and ambiguous.

Mutual Dependence and Iterated Learning

She does note, however, that attributes might be extracted from categories just as well as categories might be based on attributes. However:

Unfortunately, to state the matter in such a way is to provide no clear place at which we can enter the system as analytical scientists. What is the unit with which to start our analysis? (p. 42)

To me, this suggests a game-theoretical analysis. A category system is invented by people, but also has to be transmitted; fixed points in such an iterated learning process will be the systems that trade off difficulty of acquisition for pragmatic necessity, I guess.

This process could probably be modeled relatively easily in a multi-agent system with a set of Bayesian learners. However, such a model will probably be highly sensitive to the assumptions made about the environment of learning (e.g., the frequency of birds and the frequency of winged-ness).

Fauconnier: "Generalized integration networks" (2009)

I have quickly skimmed Gilles Fauconniers contribution to New Directions in Cognitive Linguistics (2009).

I still think he is right in pointing out that constructs like My surgeon is a butcher cannot be explained in terms of one-way mappings. I also still think that his theory doesn't have as much to offer as he himself seems to think.

Fairytale Science: Against a "Fallacy"
Somewhere in Fauconnier's mind, there must be a tiny shred of fear that his theory might be closer to psychoanalysis that to cognitive science.

This can be seen in that he explicitly takes on the following "fallacy":

A wide-ranging cognitive operation purports to explain "everything"
Corollary: Such an operation explains "too much" and is unconstrained (p. 149)

Of course, calling something a fallacy doesn't make it so. And indeed, he doesn't have any real argument that can alleviate this worry.

The closest we get is an allusion to "twelve years" of "convergent evidence" which is not followed by any reference (p. 149).

Fairytale Science: "Precise" But Not "Rigorous"
He later reiterates that conceptual integration "is a precise meaning construction operation" (p. 158), but he also admits that the analyses have to be done on a case-by-case basis:

precise analysis of the generalized networks needs to be done for any observed data, just as chemical analysis needs to be done for any unknown chemical, without adding new elements of new principles to chemistry. (p. 158)

Thus, only the skilled analyst can uncover what cognitive operations a reader must perform in order to understand a given piece of surface language.

The consequence is, he adds in a footnote, that

[...] notions like metaphor, metonymy, analogy, and counterfactual, applied to surface products, elude rigorous definition. (p. 158)

On his account, these notions are thus theory terms, embedded in his own theory.

Fairytale Science: "We Have Indeed"
Even more ridiculously, he quotes (p. 150) a single remark Feldman makes in the conclusion of his book (p. 338), calling Fauconnier and Turner's project "a bold attempt to explain much of mental life."

While "bold attempt" does not usually translate into "unqualified success," Fauconnier does not seem to have a problem reinterpreting the remark as a full-fledged endorsement of his theory:

The proper way to understand this comment is that we (Turner and Fauconnier) have indeed shown that conceptual integration plays a necessary role in human mental life as evidenced by the surface products of particular interest to humans. (p. 150)

I'm not kidding -- he actually writes this.

Showcase Semantics: A New Example
In order to demonstrate his theory, Fauconnier considers the a sentence from a news story about a new California law prohibiting smoking in certain bars. The news piece playfully combines the actual cigarette smoke with the cartoon cliché of a "smoking angry" person:

[...] hard-core smokers [...] were so angry that if they had been allowed to light up, the smoke would have been coming out of their ears. (p. 150)

This example contains two of Fauconnier's favorite phenomena: A counterfactual statements and a two-way interaction between the literal and the metaphorical level.

Showcase Semantics: Putting Pieces Together
His analysis is summarized in the figures on pp. 155 and 157. The gist is this:

1. The conventional ANGER image (or "network") is combined with a stereotypical SMOKER situation to combine an ANGRY SMOKER image which include a double (literal/metaphorical) role for the smoke.
2. The NO SMOKING and the SMOKING ALLOWED situations are combined to form a counterfactual smoking situation qualitatively different from the simple SMOKING ALLOWED situation. He calls this counterfactual the SMOKER'S ZOLOFT SPACE.
3. The ANGRY SMOKER and the SMOKER'S ZOLOFT SPACE are combined to produce the hypothetical smoke coming out of the smokers' ears.

After the Smoke Clears: Fauconnier on Fauconnier
He concludes his analysis with, if I may use my own blend, "a Rumsfeld":

Is the newspaper statement contradictory or unintelligible? Clearly, no. [...] Is the piece of data a "blend"? Clearly, yes, and indeed a visible one. (p. 157)

Shortly after that, he also assures us of the psychological reality of his speculative account (by means of a speculative argument):

Is attested data of this kind cognitively significant? Of course. (p. 158)

His argument seems to be that  because we can understand the sentence, and because he can tell a semantic story about this sentence, his story must be true (p. 158).

From Molecule to Metaphor (2006), chapters 15-17

These three chapters in Feldman's book are based on the classes he and George Lakoff give together at Berkeley. They describe Lakoff's theory of metaphor as we know it from countless other sources.

Primary Metaphor and the Basis of Meaning
Feldman begins by repeating the well-known theory that meaning grows out of first-hand experience:

For perception, action, emotions, and so on, human experience and the social relations shared by all people provide the basis for learning words. (p. 185)

[...] abstract, cultural, and technical words and concepts arise from the opulent substrate of direct experience. [...] people, as neural systems, understand abstract ideas because these concepts are mapped to and actiavte brain curcuits involved in embodied experience. (p. 185)

There is now very strong evidence that essentially all of our cultural, abstract, and theoretical concepts derive their meanings by mapping, through metaphor, to the embodied experiential concepts we explored in earlier chapters. (p. 199)

Bi-directional Metaphor and Scaffolding
Lakoff (through Feldman) makes a curious remark in the middle of chapter 16:

Once a domain of knowledge becomes well known, it can itself serve as a source domain (basis) for understanding more novel concepts. We sometimes get metaphors mapping both ways, for example, between war and sports. These create no difficulty---the appropriate concepts in each domain are activated and inferences are drawn from the combined activation. (p. 209)

And nothing more is said of that matter. This seem to be essentially Lakoff falsifying Lakoff: Saying at once that metaphors need not go "all the way down" and still claiming that they do.

More concretely, the first sentence seems to oscillate between to claims depending on which way the evidence goes: Either it means that we learn new things by comparing them to old (which everyone agrees on); or else it means that the connection between the new subject matter and the old are permanently established and remain in use (which is false).

I don't know exactly what it means that "the appropriate concepts in each domain are activated and inferences are drawn from the combined activation," and Feldman/Lakoff doesn't explain. Is the target domain known or not known? Is any information actually transferred, and if so, what?

The Invariance Principle Returns
They also repeat what Lakoff said twenty years ago:

Metaphorical mappings preserve the cognitive topology (that is, the frame and schema structure) of the source domain, in a way consistent with the inherent structure of the target domain. This is called the invariance principle. (p. 209-10)

Or, equivalently: metaphors preserve structure except when they don't. They continue:

As a consequence, the image-schematic structure of the target domain cannot be violated: one cannot find cases in which a source domain interior is mapped onto a target domain exterior, or a source domain exterior is mapped onto a target domain path. This simply does not happen. (p. 210)

But since the target domains don't have literal paths, this could hardly be more than a tautology. There would be no way to identify whether the thing we map the interior onto actually is an "interior," since no such thing is there.

The only way we could in principle falsify this claim would be if we could find some case where the metaphorical outside of the target domain could also be the metaphorical inside. Such cases do indeed exist and thus falsify Lakoff's claim. Examples are HOT/COOL IS FASHIONABLE, UPHILL/DOWNHILL IS DIFFICULT, and THICK/THIN IS IMPLAUSIBLE.

From Molecule to Metaphor (2006), chapter 10

Here's a strange quote from Feldman's book:

Even as adults, the experience we associate with a word and thus its meaning differs depending on our age, gender, profession, and so on. People who only watch a sport event or artistic performance cannot fully understand participants' conversation about the activity. (p. 130)

It's difficult to disagree with the first sentence here, and we should probably applaud that a cognitive linguist acknowledges this fact.

But it's almost equally difficult to agree with the second sentence or see any logical relation between them. Where does this extreme solipsism come from?

Perhaps this kind of thinking is the very core of the problem in cognitive metaphor theory. If everything become a matter of private experience, then our evident ability to understand each other seems like a miracle. Feldman needs a shot of Wittgenstein.

Basic Level Nonsense
Speaking of Wittgenstein, I'm still shocked at the sheer amount of nonsense associated with the term "basic level category." Elanor Rosch (in all due respect) seems to have assumed without the slightest shed of argument that all concepts come in triples like vehicle > car > truck.

That's a bizarrely unfounded claim, conceptually and empirically. What would, for instance, be the "basic level" in the following strings of inclusions?

entity  >  ( . . . )  >  artifact  >  toy  >  doll  >  puppet  >  hand puppet

As far as I can see, the only way to pin these levels onto an absolute scale with a "ground floor" would be to randomly pick one by intuition. This would essentially fold the data into the definition and empty the theory completely of any meaning.

From Molecule to Metaphor (2006), chapter 7

As a part of his general introduction to non-neurological psychology, Jerome Feldman cites a 1979 study that indirectly sheds some new light on Tim Rohrer's fMRI findings.

Ambiguity and Primed Priming
In the 1979 priming experiment by Tanenhaus et al., subjects were asked to decide whether a given string (e.g., season) was an English word. They did this in one of two conditions.

In one condition, the subjects were primed with a completely unrelated sentence, e.g., she was afraid to talk. In the other condition, they were also primed with an unrelated sentence; but in this case, the sentence contained an ambiguous word, e.g., she was afraid to fall.

These sentences were constructed so that there was only one possible way to read the ambiguous word. For instance, in the example above, the word fall can only be read as the verb, not the noun. Any activation in the concept of falling would be a secondary effect and would not contribute to the understanding of the sentence.

However, the researchers consistently found that the indirectly related primes would nudge the subjects into faster and more accurate decisions, just as if they were responding to unproblematically related words (e.g., cat--kitten).

It thus seems that the seasonal meaning of fall is, to some extent, activated by the sentence she was afraid to fall. This is not plausibly explained as a metaphorical process, especially since a number of the ambiguous words were clearly not related, e.g. I put it in the box as a prime for fight.

Relation or Disambiguation?
Now look again at Rohrer's findings: He had his subjects read sentences that contained metaphorical uses of words such as grasp or fist, and he found that such words, on average, lead to increase in activity in same the brain regions that are associated with the actual hands and arms.

He took this as a confirmation of the conceptual mapping hypothesis. Since hand metaphors activate the hand concept, literal meaning must be involved in metaphorical understanding. His findings are indeed consistent with this hypothesis.

However, as the experiment by Tanenhaus et al. shows, the brain activity could just as well be a byproduct of a disambiguation process. We can take grasp the theory and grasp the cup to apply two distinct meanings of the word grasp. Both meanings would then presumably be activated during reading, just like the verb and noun meanings of box are both activated during the reading of I put it in the box.

Rohrer's results are then equally also consistent with another theory that does not assume a productive and active link between the two meanings of grasp. Instead, it assumes a link from the sound grasp to two meanings, but only one is picked on the basis of the context.