In response to my posts about Rico (the border collie who can fetch in response to about 200 different spoken words), Mark Seidenberg wrote to draw my attention to a paper that he co-authored nearly 20 years ago:
For me, Rico is deja vu all over again, since the discussion so closely recapitulates responses to the early reports about Kanzi, the bonobo, who was also supposed to be an incredibly fast learner, have a remarkable vocabulary, etc. We even used the Helen Keller case in a 1987 article about Kanzi.
(Seidenberg, M.S., & Petitto, L.A. (1987). " Communication, symbolic communication, and language." Journal of Experimental Psychology: General, 116, 279-287.)
Mark's note agrees with the interpretation of the Helen Keller "water insight" story that I took from Walker Percy:
I think it's highly doubtful that Helen Keller's memory of this event is accurate, or that there is a naming-insight-moment that occurs in normal child development. However, her description (and the dramatization of it in The Miracle Worker) does effectively bring out the idea that there might be something about having the concept of a name, rather than mere associations between linguistic forms and objects (or even classes of objects).
Kerim Friedman also sent email on this point, citing his post last fall at Keywords that quotes Keller's autobiography (following a comment by Lisa at Language Hat's site) in support not only of general pre-disease language acquisition, but even of specific knowledge of the word "water":
I am told that while I was still in long dresses I showed many signs of an eager, self-asserting disposition. Everything that I saw other people do I insisted upon imitating. At six months I could pipe out "How d'ye," and one day I attracted every one's attention by saying "Tea, tea, tea" quite plainly. Even after my illness I remembered one of the words I had learned in these early months. It was the word "water," and I continued to make some sound for that word after all other speech was lost. I ceased making the sound "wah-wah" only when I learned to spell the word.
This further undermines the mythic image of a moment of electric insight -- it becomes more like Plato's notion of learning as remembering -- but as Mark says, the main point of the story is that naming is more than associating signs with objects. In Walker Percy's way of talking, the key difference is between a symbol and a sign. I quoted Percy's 1975 essay, which in turn quoted Langer's 1957 book and various even earlier sources, because I wanted to avoid triggering the understandable frustration that animal communications researchers feel when they accuse skeptics of "moving the goalposts" after each new demonstration of animal abilities. The 1987 Seidenberg and Petitto paper also helps to make it clear that these issues have been out there for some time.
Mark's email continues:
Most critics of the animal language research (over what is now a 30+ year period) have tended to grant that the animals in question had acquired lexical knowledge comparable to a young child's, but drew the line at syntax. Whereas I took the position that the animals couldn't even name; if you look at their behavior closely it deviates from what you would expect if the animal really knew what a name is. And if they don't know what names are, how could that be taught?
In the 1987 paper, Seidenberg and Petitto wrote that "intriguing results from an ongoing study of two pygmy chimpanzees" suggest that "their behaviors are similar to nonlinguistic gestures used by 9- to 16-month-old children". They argue that to characterize animals' behaviors "in terms of a general notion of symbolic communication ... is unsatisfactory because the term is not clearly defined and it is not clear what range of behaviors it subsumes. Terms such as gesture, symbol, and word ... are not equivalent."
The crux of their conclusion is that when chimpanzees use signs in natural contexts, they "mand" in the sense of Skinner (1957)[see below]. They argue that "[t]his behavior is surely communicative; the important point is that it does not require knowledge of words or symbols at all. Under the appropriate circumstances, a smile can serve the same function."
Putting it another way, they write:
Our view is that Kanzi's behaviors are more like the use of tools than the human use of language. Tools are the instruments by which we attain certain outcomes. They are not symbols.
Those interested in philosophy of language will note that this is similar to Searle's Chinese Room argument, which they cite.
Mark's recent email points out that
The fact that Rico is always comprehending language in the context of fetching is related to our observation that Kanzi (and other chimpanzees like Nim Chimpsky, with whom I worked for a while) used their "language" for instrumental purposes like getting food. I would also note how notoriously difficult it is to assess comprehension in children, let alone dogs. How much information does the child/animal need to have in order to perform a task (like picking out the referent of a word)? Often such tasks can be done on the basis of very partial knowledge that would not necessarily license the attribution that they know the words.
Rico doesn't change anything about the linguistic capacities of animals; however, his use of mutual exclusivity ("process of elimination" is how they are describing it in the news reports) to determine the likely associate of a new word makes him a very smart fellow.
Specifically, they suggest that
Kanzi has learned about the instrumental functions of lexigrams in the experimental context. He does not know that lexigrams designate, represent, symbolize, or name objects and events; rather, he knows how to use them in order to effect desired outcomes such as obtaining objects, being allowed to engage in favored activities, or receiving the approval of his trainers. Lexigrams are the means by which he obtains positive responses from the teachers who control these outcomes. This account explains several aspects of the chimpanzee's performance that would otherwise seem incidental or unrelated.
The origin of this hypothesis was their own work in the late 1970s with another chimpanzee, Nim, from which they concluded that "he seemed to know the outcomes associated with producing signs--that is, their pragmatic functions--not the concepts associated with them, or what they named, or that they were names at all".
One of the reasons for this conclusion is the difference between behavior in "vocabulary tests" and in "naturalistic exchanges".
The vocabulary test shows that Kanzi can associate lexigrams with pictures of objects and spoken words; conversely, he can associate spoken words with lexigrams and pictures. This behavior is remarkable, and the fact that he learns quickly without the procedures used in previous studies is important. The difficult question is whether this behavior indicates that lexigrams function as symbols or names. The same question applies to the similar, though more limited, behavior observed in pigeons, who are capable of learning to associate an arbitrary response with exemplars of categories such as trees or bodies of water. It also applies to the early communicative behavior of children.
In the naturalistic exchanges, Kanzi used lexigrams much more broadly:
Kanzi knows something about the outcomes associated with lexigrams and uses them--communicatively--to effect these outcomes. Thus, he has learned to produce juice in contexts where his trainers will interpret it as appropriate, thereby facilitating outcomes such as receipt of juice or a trip to the juice location, not because juice designates specific objects.
Those who think that the chimps have learned words are prone to interpret this broadening as something like metonymy (though Seidenberg and Petitto do not use that word):
For example, if Kanzi learns to touch the symbol for strawberries when he wants to travel to the place where they are found, when he is asking for one to eat, and when shown a photograph of strawberries, he will probably extract the one common referent (red sweet berries) from all of those different circumstances, and assign to that referent the symbol, strawberries. (from Savage-Rumbaugh et al. 1986)
But as Seidenberg and Petitto argue,
Whatever the plausibility of this assumption, it requires empirical validation. We agree that in producing strawberry Kanzi recognized something common to all the situations in which it is used, but what? Is the common element strawberries or the fact that these are the situations in which strawberry can be used with a positive result?
S & P suggest one type of empirical test:
In experiments by Markman and Hutchinson (1984), a child was presented with a picture of a target object (such as cow), a taxonomically related alternative (pig), and a thematically related alternative (milk). Thematically related items were causally or temporally related to the target; other examples include door/key (object/instrument) and ring/hand (object/location). The experimenters introduced a novel word for the target object (e.g., sud). The child's task was to choose the picture that represented another sud ("See this? It is a sud. Find another sud that is the same as this sud."). On approximately 80% of the trials, children chose the taxonomically related alternative rather than the thematically related alternative. Hence, their initial hypothesis about an unfamiliar word is that it refers to a category of objects rather than objects that happen to be causally or temporally related.
In using a lexigram in reference to an object such as juice and to the location where it is found, Kanzi was apparently responding on the basis of temporal and causal relations between these entities.
S & P also observe that
It is possible to extract from child language corpora utterances that resemble Kanzi's.
For example, a child is observed to consistently use the word daddy in reference to her father and no other person. The child then sees the father's special easy chair when he is not present, points to it, and says "daddy." Therefore, the child is using daddy in reference both to a person and to the location where he is often found, much like Kanzi used juice in reference to a drink and its location. The comparison between child and chimpanzee turns on whether the same processes and types of knowledge underlie their respective utterances, something that cannot be determined by examining individual utterances. Our claim is that the bases of these utterances are in fact very different.
Read the paper for the rest of the argument. Two lessons of the long history of research in this area:
Still to come: a close look at the Rico article, and at Paul Bloom's discussion of it.
* B.F. Skinner proposed categorizing instances of speech according to how they are reinforced -- see this description. Attempting a brief summary... his categories were echoic, mand, tact, interverbal and autoclitic. An echoic utterance is a repetition of another's pattern ("Cookie" in response to "Can you say cookie?); a mand is a demand or directive ("Cookie" in order to get a cookie); a tact is an utterance that provides information rather than attending to a state of deprivation ("Cookie" to name a picture of a cookie); an interverbal is a discourse-related word like "please", or a purely associative conversational sequence (you say "milk", I say "cookie"); autoclitic speech seems to be internally-directed speech, or perhaps everything left out of the other four categories -- namely just about all linguistic behavior?
[Update 6/16/2004: interesting comments, and informal experiments with another dog named "Rico", at The Panda's Thumb.]Posted by Mark Liberman at June 15, 2004 09:42 AM