Please post your comments on classes and readings for 9/17 and 9/23 here.
7 thoughts on “Artificial phonology and concept learning”
Monica Bennett
There were a variety of languages of which the participants were native speakers in the experiments mentioned in these readings. How much does the native language of the participants matter to artificial phonological learning? It seems like you would get some interference from the participants’ native phonotactics. This clearly wouldn’t be a confound if your sample all has the same language background, but I’m wondering which elements of the artificial phonologies would be more difficult to learn depending on it.
Monica – this is a great question. There’s a bit of discussion of this in the second of the Moreton and Pater 2012 papers, but let’s make sure we talk about it in class.
When doing the two chapter readings, I noticed that the domain specific/domain general debate started to pop up. I was curious if there is a certain camp that phonologists seem to side with, or if it’s as equally divided as it is in developmental psychology.
Building on that, I was hoping to understand why substantive biases are by nature domain specific (I think I’m still having a little trouble getting the differences between substantive and structural).
Ashley – good questions. I think that the general view, at least in generative linguistics as a whole, is that “language is special”, and that the learning mechanisms would be specific to language. But most of the debate about this has to do with syntax, rather than phonology, and I wouldn’t be surprised to find that phonologists are more receptive to the notion that learning mechanisms might be shared across domains.
Substantive biases are domain specific insofar as they are tied to articulatory or perceptual ease in language, or properties of language-specific representations. But now that I think about it, one could potentially imagine substantive biases that would be tied to perceptual factors that go across domains.
I, too, am still trying to wrap my head around the distinction between substantive and structural biases. Can one make the distinction that substantive biases are biases toward patterns that serve the function of language (i.e., facilitate the encoding and transmission of information), whereas structural biases are biases that are specific to a certain language, but do not serve any function? One related question I have is whether across languages, we see relatively the same number of phonotactic constraints (not the same constraints, but the same number and diversity of constraints). Or, do we find that some languages are heavily phonotactically constrained and others are much more diverse in the patterns that are allowed? If the latter is true, then one can ask whether the highly constrained languages are somehow deficient as a medium for communication (i.e., are somehow not able to encode as rich a variety of semantic representations). Are these languages poorer vehicles for communication? If this is not the case (and I doubt that it is the case) can’t we conclude that the diversity of legal patterns seen in other languages serve no underlying purpose? Are these “structural biases†in the sense that they are arbitrary in regard to the function of language?
I’m wondering about the use of ANOVAs in Kurtz et al. Since we’re looking at a binary response variable, should we be using mixed effects logistic regression? I’d be curious if anyone knows why an ANOVA might be better suited to this case.
See Jaeger, T. F. 2008. Categorical Data Analysis: Away from ANOVAs (transformation or not) and towards Logit Mixed Models. Journal of Memory and Language 59, 434-446 available here: http://www.bcs.rochester.edu/people/fjaeger/
Kurtz et al’s conclusions that type II pattern is learned faster/better if the learners postulate a concrete rule made me think about language acquisition vs language learning (as in L2). Is there evidence that the L2 learners might acquire a phonological structure (vowel harmony maybe?) which can be formalized in words faster than the native speakers of this language (eg. vowel harmony acquired relatively late in the process of natural language acquisition, but learned relatively easy by L2 learners of this language, given a rule how it works). I bring vowel harmony here as an example of a phonological phenomenon, not because its acquisition works the way I described here.
There were a variety of languages of which the participants were native speakers in the experiments mentioned in these readings. How much does the native language of the participants matter to artificial phonological learning? It seems like you would get some interference from the participants’ native phonotactics. This clearly wouldn’t be a confound if your sample all has the same language background, but I’m wondering which elements of the artificial phonologies would be more difficult to learn depending on it.
Monica – this is a great question. There’s a bit of discussion of this in the second of the Moreton and Pater 2012 papers, but let’s make sure we talk about it in class.
When doing the two chapter readings, I noticed that the domain specific/domain general debate started to pop up. I was curious if there is a certain camp that phonologists seem to side with, or if it’s as equally divided as it is in developmental psychology.
Building on that, I was hoping to understand why substantive biases are by nature domain specific (I think I’m still having a little trouble getting the differences between substantive and structural).
Ashley – good questions. I think that the general view, at least in generative linguistics as a whole, is that “language is special”, and that the learning mechanisms would be specific to language. But most of the debate about this has to do with syntax, rather than phonology, and I wouldn’t be surprised to find that phonologists are more receptive to the notion that learning mechanisms might be shared across domains.
Substantive biases are domain specific insofar as they are tied to articulatory or perceptual ease in language, or properties of language-specific representations. But now that I think about it, one could potentially imagine substantive biases that would be tied to perceptual factors that go across domains.
I, too, am still trying to wrap my head around the distinction between substantive and structural biases. Can one make the distinction that substantive biases are biases toward patterns that serve the function of language (i.e., facilitate the encoding and transmission of information), whereas structural biases are biases that are specific to a certain language, but do not serve any function? One related question I have is whether across languages, we see relatively the same number of phonotactic constraints (not the same constraints, but the same number and diversity of constraints). Or, do we find that some languages are heavily phonotactically constrained and others are much more diverse in the patterns that are allowed? If the latter is true, then one can ask whether the highly constrained languages are somehow deficient as a medium for communication (i.e., are somehow not able to encode as rich a variety of semantic representations). Are these languages poorer vehicles for communication? If this is not the case (and I doubt that it is the case) can’t we conclude that the diversity of legal patterns seen in other languages serve no underlying purpose? Are these “structural biases†in the sense that they are arbitrary in regard to the function of language?
I’m wondering about the use of ANOVAs in Kurtz et al. Since we’re looking at a binary response variable, should we be using mixed effects logistic regression? I’d be curious if anyone knows why an ANOVA might be better suited to this case.
See Jaeger, T. F. 2008. Categorical Data Analysis: Away from ANOVAs (transformation or not) and towards Logit Mixed Models. Journal of Memory and Language 59, 434-446 available here: http://www.bcs.rochester.edu/people/fjaeger/
Kurtz et al’s conclusions that type II pattern is learned faster/better if the learners postulate a concrete rule made me think about language acquisition vs language learning (as in L2). Is there evidence that the L2 learners might acquire a phonological structure (vowel harmony maybe?) which can be formalized in words faster than the native speakers of this language (eg. vowel harmony acquired relatively late in the process of natural language acquisition, but learned relatively easy by L2 learners of this language, given a rule how it works). I bring vowel harmony here as an example of a phonological phenomenon, not because its acquisition works the way I described here.