Nov 17 ====== Mendoza/etal:2003 ----------------- Page 12: What is the definition of the "categorical rule"? Page 12: What is a "variable rule"? Comments: I actually found the paper pretty informative, though I would have preferred if it explained categorical and variable rule more closely in the beginning. Another problem for me was the lack of background knowledge of VARBRUL, while trying to understand the results of the analysis. I found especially interesting the use of interlocutor-free speech episode as a controlling measure in the natural speech recordings. Since there was a ceiling effect for /ay/ monophthongization in Winfrey's speech when referring to a 'zero'-referent as opposed to African American (AA) referents, for example, what other variables could have interacted with the AA referent for this to be the case? I suppose gender as a variable being significant only between 'other' and specified gender categories correlates with this ceiling effect. What about new methods for collecting data nowadays, like crowdsourcing, or those websites where you can record a sample of your dialect and upload it to a database? Isn't the data then a bit "unnatural"? Because even though the participants are in an environment they are comfortable in (presumably), they are still aware they are recording an experiment and may not produce the most natural speech. Are there other (simpler) ways of calculating probabilities for variable rules? Have other models, more recent, that have successfully mitigated some of VARBRUL's limitations? In terms of the Winfrey samples, I wonder if the samples are good for testing style shift, because even if Winfrey wasn't talking to a specific interviewer, isn't talking into a camera an unnaturalistic situation? Pierrehumbert:2006a ------------------- Page 3: In which way do models based on network structures produce counter-intuitive behaviors? Page 10: In the subchapter "Dissociations between perception and production" what kind of studies or experiments could help explain the details? Comments: I think it provided a good analysis of the properties of VARBRUL as well as a wide overview over the field itself. I found it also very well organized in how it first talked about its general properties, before really weighing its strengths and weaknesses and comparing it with other tools. Frequency bias in perception implies that infrequent phonemes only survive in robustly distinct from their competitors - so how do frequent phonemes that are closely related survive in terms of perception; is this even something that has been observed? How can external social factors be included or observed in exemplar theory models? Might be too soon to say, but it feels the relationship between forensic phonetics and sociolinguistics is close but not quite mentioned in the papers we read so far. Why could that be? General ------- Different speech communities favor different phonetic contexts (especially regarding the following context) for the application of monophthongization Frequency effects are also dependent on the speech community: word frequency has a much bigger effect for non-African-American referees than for African-American referees. Frequent words are therefore not more prone to monophthongization for African-American speakers. Could you elaborate on the discussion or dispute on whether particular patterns of variations are categorical or gradient? How does age change the vocal tract apparatus? Is it simply shortened? What are the reasons why words can be learned significantly better than new speakers? Does the linguistic variable simply describe the fact that speakers have different possibilities of what they could say and the decision point is the linguistic variable, i.e. the influence of the environment in which the speaker utters a sentence? What exactly are the differences between patterns of variation and the architecture of variation? What would be the best way to minimize the effects of lab and interviewer as much as possible, but still collect enough data from underrepresented linguistic data, since certain variants may be rare in naturalistic discourse? It is interesting to note that speakers align their language by reusing language patterns. I think it is also worth noting that a relatively high F0 score for women is (also) socially anchored. It would be interesting to investigate the reasons for this anchoring.