Dear Students!
In the next seminar on the 18th of February we will have a fun and extremely knowledgeable speaker on the topic of behavioral Evo-Devo - Bob McMurray!
Please read the following paper for his seminar:
Pigeons acquire multiple categories in parallel via associative learning: A parallel to human word learning?(2014)
Edward A. Wasserman a,b, Daniel I. Brooks c, Bob McMurray
Don't forget, your blog entries are due by 9am on Wednesday before the seminar!!!
Have fun!
Julia
Overall, this paper was quite interesting.
ReplyDeleteI found the route back to behaviorism to be a good move. As Darwin mentions in a section of "The Descent of Man", if two groups of entities have similar behaviors, it seems parsimonious to assume similar mental capacities between them. So this way, the behaviorist route can still retain cognitivist claims, though it only studies the observable behaviors.
Also, I felt as though they took great strides to attempt to make their experiment very similar to how children would learn, which is quite important.
Also, I'm not sure if the "category vs rote memorization problem" is really that much of an issue (not to mention, as the authors point out, categorization is a MUCH more parsimonious account). Sure it is if we want to know if they obtained an understanding of categories, but proposing that the birds memorized 128 stimuli and the correct responses is equally praiseworthy and forces us to attribute much in the way of memory to them. (though the novel stimuli section gives much more credibility to the idea that they were categorizing). However, I was a bit confused by the results they found in section 3.2.1.
Lastly, the role of association is making a huge resurgence in much of cog sci, it seems. For one, it largely coheres with the connectionist network view of the mind, as well as dynamical systems theory. Recently, it has been proposed that mirror neurons are nothing more than association machines. All-in-all, association seems to be a very powerful, and until now under-appreciated, tool that can be scaled up to explain a lot of behavior.
I enjoyed this paper. It was interesting to read a behaviorist study using this type of modeling in nonhuman animals. This study adds support to the idea that children's word learning could be attributed in part to simple associative learning. Such an association could be created between the sound and the idea of an object.
ReplyDeleteFrom what I gathered, this article touches on how associative learning in nonhuman animals shows both similarities and dissimilarities with humans. However, it does not appear to clearly state what stage of language development would reflect this behavior. I was confused as to whether this was simply being held as a representative model across initial acquisition (up to age 2) or does this extend past the vocabulary spurt.
So essentially, is this associative learning mechanism only in the early, more initial stages of word learning or is it more working in parallel with later referential word learning mechanisms as the child matures?
I'm curious but it seems most plausible that these two are working together in human word learning and I think the authors would agree. Associative learning appears to be a building block of sorts that could later add to more referential mechanisms. Learning more about associative processes will lend us better insight into the language learning process. Neural network studies assessing the slow start of language through the vocabulary spurt in computational models show that there is no change in the learning mechanism during this shift. Rather, as the network gets stronger, acquiring more words and mapping referents, the process becomes faster as well. As the network acquires knowledge, it then uses the prior experiences to process new information. This process implies that associative learning is important to the both the rate and process of acquisition. I am interested in better understanding this type of model.
This article offers an in-depth overview of a longstanding debate which, in many ways, dwells at the roots of cognitive science. By addressing the questions "Are complex - perhaps uniquely human - behaviors like language acquired via specialized and perhaps innately constrained cognitive mechanisms? Or can such behaviors emerge from more basic and general mechanisms like associated learning?" (99), the authors provide evidence in favor of the latter claim with a new biological model that may help adjudicate several core issues of the debate, along with addressing these issues directly throughout the paper.
ReplyDeleteIn light of this framework, I have several questions concerning section 4.3: Association vs. logical inference. The authors here suggest two primary reasons why the pigeons in the study were learning via association instead of a richer cognitive process. First, the nature of the task at hand seems ill-suited for an inferentialist account. Second, after properly adjusting the learning curve, we find that it does not fit the "sudden learning" hypothesis of Gallistel et. al. (2004). The authors also note that there is a distinction between the contrasting symbolic process approach in question and a Bayesian approach. I was wondering in what ways should we understand the relationship between Bayesianism and the alternative choices of associationist learning and logical inference? Does evidence for the Bayesian hypothesis offer support for the associationist framework or are the two categories related but logically distinct?
When discussing models, the authors talk about a study that was done using a “computational instantiation of the Rescorla-Wagner learning rule” and I’m not sure what that is, but they say that the study showed that adults tended to use a logical or inferential approach when “inferring the name of a novel object” (101).
ReplyDeleteFirst, what is this learning rule?
Second, and I think this may follow Megan’s question, but the study referenced (on 101 in regards to the learning rule) seems to suggest that adults begin to favor an inferential process at some point. Then later in the paper it says that adults can learn 16 categories in an hour 116). I’m wondering about the age group that this study is analogous to, and is there any information on when and why this transition away from associative learning happens, or is the preference in adults dependent on the type (or amount) of learning that needs to be done?
First, I do not understand how the author draw the Fig. 3. Second, after reading this paper, I feel the way, which the human let the Pigeon to understand the words and pics, is similar to the human make machine learning. Therefore, there may be a question that, altought pigeon is alive, is it possible that the pigeon uses different ways for the Active learning and Passive learning (maybe learning the human words is a kind of Passive learning for them).
ReplyDeleteThe purpose of this research seems to illustrate, through observing pigeons, whether biological system can learn a many-to-many set of categorization, and try to link it to human word learning under the associative learning aspect. It was quiet complicated to me since authors in this paper deal with diverse aspects, but I like such a way of carefully analyzing the data to avoid misinterpretation of observation from an animal study.
ReplyDeleteMy question is that this study was designed in a way that birds are learned and categorize words with visual stimuli. However, word learning beginning at early age is carried out, in most cases, combined with auditory information (visual stimuli/auditory feedback or vice versa). I don’t think this would not significantly affect the results, but it could make more sense to use auditory stimuli when we consider the way of word learning in infants.
My first question is that why the 128 photographs were grayscale and the 16 pexigrams were with colors?
ReplyDeleteDid the pigeons make errors among the photos from different categories but with similar shapes? For example, some hats may look like cakes (rounded), phones (smart phones) may look like crackers, or trees may look like flowers, especially when they were presented with black and white photos. I could not get access to the link on page 105 “online supplement (S1)” to see if the problem exists.
In contrast, mutual exclusivity (e.g., only call the duck in the bathroom "duck" but not other ducks) might also be present in pigeons, and the authors pointed out on page 119 that this phenomenon may not be obvious. But I think that after a certain amount of presentation, the pigeons should have overcome (1) shape bias and (2) mutual exclusivity to categorize these photos correctly.
The article begins by framing the debate on language between the poles of Chomsky and Skinner; that is, does language emerge because of specialized and perhaps uniquely human neural and anatomical structures, or are we just situated somewhere on a spectrum of association forming capabilities. The article seems to split the difference between the two, suggesting that while inferential learning is certainly the MO of adult human learning, stimulus/response associations might play a much larger role in early language learning than the nativists might have thought.
ReplyDeleteHowever, I think that there’s a third figure that can explain these results in a more fruitful way that just splitting the difference between associative behaviorism and cognitive modularity (and he is a figure that fits in far better with the theoretical thrust of our seminar): Lev Vygotsky. The article seems to be more or less stuck looking at what’s going on in the organisms head, when it might be more interpretively value to look at what’s going on with the situation. Aren’t we placing these pigeons in a stimulus rich experimental apparatus and guiding them through participation (to use Barbara Rogoff’s language) in the sort of linguistic categorization practices that we use to navigate the world and communicate with one another? More than this points to any conclusions about whether an individual learns language by association or some innate cognitive roadmap, it really seems to point us in the direction of Vygotsky’s idea of the “zone of proximal development.” This article seems to be demonstrating here, whether they realize it or not, that regardless as to what’s going on ‘in the head’ of children or pigeons, being guided through participation in community language practices really seems to be the sine qua non of language development.
Wasserman, Brooks and McMurray address a very large question relating to psychology, linguistics, and theories of human development: Who is correct, Skinner or Chomsky?
ReplyDeleteVery succinctly, the Skinnerian view, termed behaviorism, is that learning takes place as a result of consequences and rewards for behavior. Positive stimuli reinforce concurrent behaviors, while negative stimuli deter behaviors. The Chomskyan view is that the ability to learn language is innate in humans; it may perhaps be compared with a software system ready to decode vocabulary and grammar.
The underlying questions related to Chomsky and Skinner may be restated in a number of ways, including: How active is the human mind in the language learning process? Is the mind a blank slate for language, as presumably for other learning tasks, or does it come partially pre-programmed?
Another layer of questioning addressed by Wasserman et al. involves the degree of similarity between humans and other species. It is intriguing to note that while studying pigeons directly, Wasserman et al. aim the focus of their research at human language learning. This may be considered approaching the topic from a side angle, and involves some ingenuity.
The Wasserman et al. article includes vocabulary used in specific ways with regard to this academic discussion, some of it unfamiliar to me. ‘Emergentist’ is one such term. I look forward to hearing from McMurray in person about the terminology, theories, and other particulars of this research.