August 23, 2006
Analogical modeling
The motto of my old Japanese professor, Watabe Sensei, when he was consulting for a TESOL software company I used to work for (now sadly defunct) was: "Examples, examples, examples!" The problem was, our Japanese clients insisted on: "Grammar, grammar, grammar!" And the customer is always right, even when they're wrong (especially when they're paying the bills).
Another linguistics professor from my alma mater, Royal Skousen, has been working on the theory to back up Watabe Sensei's insistence on more examples. Skousen calls his theory "Analogical modeling." He argues that language acquisition is based on pattern recognition and substitution, not rules. That is, we model language by forming analogies, not by applying grammar rules:
When acquiring and using a new language structure, then, we look for the largest number of examples that are close to the context with which we are concerned, systematically eliminating all those that fail to match the specific context in favor of those closest to the mark. So if C kind of resembles A, but B more closely resembles A, reject C in favor of B.
In other words, the only rule is that the closest match wins, and the more close matches the better.
Skousen posits that even when looking for close matches, trying to work down through a binary tree matching individual elements is ultimately less efficient than simply matching the entire pattern. "Instead of dealing with probabilities, it [is] much simpler to directly store the examples." This is true as well in computer programming. Rather than algorithmically generating numbers like PI, it is far more efficient simply to store them as static variables.
This confirms my own experience that when it comes to comprehending complex phrases, example-based dictionaries like Eijirou are far more useful than grammar texts. In the case of Eijirou, as Skousen would put it, "The usage is the description."
Here are two real-world applications. Describing how he studies Japanese, Chiba Lotte Marines manager Bobby Valentine refers to a course called Step Up Nihongo, devised by Shigekatsu Yamauchi. I can't vouch for the pedagogy of the program, but a Yomiuri Shimbun article about Valentine stresses that "SUN employs a lot of pattern drills, as Yamauchi believes mastering the patterns is the best way to rapidly learn Japanese."
And one Professor Waragai, at the Tokyo Institute of Technology, states that in order to master German sufficiently to gain admittance to the University of Vienna:
This sounds an awful lot like what I did at the MTC for eight weeks when I was first studying Japanese. The danger, of course, is that the "patterns" and "examples" will fall into the swamp of audiolingualism, and be hijacked into the teaching of abstract grammatical forms. The patterns and examples should always come from real-world applications and have pragmatic and immediate use in the real world.
Although at the time the curriculum did rely a great deal on audiolingualism, this is one thing that the MTC did right. We memorized a lot of material that was "real" Japanese, not simplified, dumbed down, or otherwise linguistically compromised. According to this formulation, the difficulty of the grammar is never an issue, only the job the language is intended to accomplish.
Of course, memorizing those 500 examples as Professor Waragai did does not mean he was then "fluent." Even after eight weeks at the MTC, when I got to Japan I still didn't understand a thing. But what it does do is take the training wheels off and give you a driver's permit. No tentatively dipping your toe in the water--you feel emboldened to jump right in.
Another linguistics professor from my alma mater, Royal Skousen, has been working on the theory to back up Watabe Sensei's insistence on more examples. Skousen calls his theory "Analogical modeling." He argues that language acquisition is based on pattern recognition and substitution, not rules. That is, we model language by forming analogies, not by applying grammar rules:
No training stage occurs in [Analogical modeling], except in the trivial sense that one must collect exemplars [examples] in order to make predictions. There is no setting of parameters nor any prior determination of variable significance. The significance of any combination of variables is always determined in terms of the given context for which we seek a predicted outcome. Each prediction is done on the fly.
When acquiring and using a new language structure, then, we look for the largest number of examples that are close to the context with which we are concerned, systematically eliminating all those that fail to match the specific context in favor of those closest to the mark. So if C kind of resembles A, but B more closely resembles A, reject C in favor of B.
In other words, the only rule is that the closest match wins, and the more close matches the better.
Skousen posits that even when looking for close matches, trying to work down through a binary tree matching individual elements is ultimately less efficient than simply matching the entire pattern. "Instead of dealing with probabilities, it [is] much simpler to directly store the examples." This is true as well in computer programming. Rather than algorithmically generating numbers like PI, it is far more efficient simply to store them as static variables.
This confirms my own experience that when it comes to comprehending complex phrases, example-based dictionaries like Eijirou are far more useful than grammar texts. In the case of Eijirou, as Skousen would put it, "The usage is the description."
Here are two real-world applications. Describing how he studies Japanese, Chiba Lotte Marines manager Bobby Valentine refers to a course called Step Up Nihongo, devised by Shigekatsu Yamauchi. I can't vouch for the pedagogy of the program, but a Yomiuri Shimbun article about Valentine stresses that "SUN employs a lot of pattern drills, as Yamauchi believes mastering the patterns is the best way to rapidly learn Japanese."
And one Professor Waragai, at the Tokyo Institute of Technology, states that in order to master German sufficiently to gain admittance to the University of Vienna:
1. Neither to see nor talk with anybody during this three weeks.
2. Study German till morning to night.
3. Memorize 500 examples
This sounds an awful lot like what I did at the MTC for eight weeks when I was first studying Japanese. The danger, of course, is that the "patterns" and "examples" will fall into the swamp of audiolingualism, and be hijacked into the teaching of abstract grammatical forms. The patterns and examples should always come from real-world applications and have pragmatic and immediate use in the real world.
Although at the time the curriculum did rely a great deal on audiolingualism, this is one thing that the MTC did right. We memorized a lot of material that was "real" Japanese, not simplified, dumbed down, or otherwise linguistically compromised. According to this formulation, the difficulty of the grammar is never an issue, only the job the language is intended to accomplish.
Of course, memorizing those 500 examples as Professor Waragai did does not mean he was then "fluent." Even after eight weeks at the MTC, when I got to Japan I still didn't understand a thing. But what it does do is take the training wheels off and give you a driver's permit. No tentatively dipping your toe in the water--you feel emboldened to jump right in.
Labels: language
Comments