From Joe Pater
This year is a good candidate for the 25th anniversary of Optimality Theory. 1993 marks not only the publication date of Prince and Smolensky’s book as a technical report, but also the year of the establishment of the Rutgers Optimality Archive (tell that to your ArXiv-loving CS friends!).
This seems like a good occasion to reflect and reminisce, so I thought I’d start this discussion thread. I had an earlier chance to reflect on the history and status of violable constraints in phonology at the 2015 mfm “Whither OT” fringe workshop. My handout, and some related discussion, can be found here: https://blogs.umass.edu/comphon/2015/05/29/discussion-whither-ot-handout/. At the same date as that workshop, a celebration of Alan Prince’s career was held at Rutgers, at which he was presented with the Short ‘Schrift (Baković ed.): https://princeshortschrift.wordpress.com. This multimedia volume contains lots of relevant material to browse. Marc van Oostendorp has recently circulated his history of OT. The story of the beginning of Prince and Smolensky’s collaboration is told by the protagonists here: https://blogs.umass.edu/brain-wars/the-debates/prince-vs-smolensky/. Finally, the handouts from Prince and Smolensky’s 1991 Linguistic Institute course, which include the 1991 Arizona Phonology conference handout “Optimality” (another candidate birthday) are here: http://www.cs.colorado.edu/department/publications/reports/docs/CU-CS-533-91.pdf.
I don’t have any more serious reflections or general historical notes to add at this point, so I’ll just add a few reminiscences. I was a PhD student at McGill in 1993 when Prince and Smolensky’s manuscript appeared in Glyne Piggott’s mailbox. We made bound copies at the copy shop, and got reading. I have a crystal clear memory of reading it on the shore of Loughborough Lake that summer, thinking “this isn’t how everyone says phonology works”. I was skeptical for a while, especially about the claim that there were no rules (what was Gen, after all?), and it did take a lot of work to understand any of it. But I ended up being an early adopter, mostly because it allowed me to formalize some ideas I had about English stress.
I had the great fortune of being accepted as an alternate at the 1994 Prosodic Morphology conference in Utrecht, where John McCarthy nicely invited me to come sit in on his seminar in Correspondence Theory at UMass (I wish the Amtrak Vermonter still went all the way to Montreal!). I thus got to spend a lot of time interacting both with people working in OT, as well as with OT-skeptics (many Canadian phonologists, including my advisors at that time), which I think was a really good learning experience.
Another crisp memory of the early OT days is the “Is the Best Good Enough?” conference at MIT. Despite the somewhat tendentious title, the call also emphasized potential common ground between OT and Minimalism, which at that point included notions like minimization of derivational length. In his talk, Chomsky abandoned that common ground, claiming that comparison of derivations was problematic on computational grounds. I wish I could remember the exact phrase – it was something like “obviously computationally intractable”. In the question period, Smolensky asked him to elaborate, citing his own work with Tesar on computing OT. As I recall, Chomsky said something that didn’t answer the question.
Jason Shaw (Yale) and I are interested in how languages syllabify the consonant clusters after V1 is deleted in C1V1C2V2 configuration, especially word-initially. There are two strategies:
C1V1.C2V2 => C1C2V2
(2) C1 keeps its syllabicity
C1V1.C2V2 => C1.C2V2
Our Facebook-based search, to our surprise, showed that there are more studies that argue for (2) than (1). For example, in English the word initial schwa in “support” can delete, yielding [s.phort]. We suspect that [s] is not resyllabified because the following [p] is aspirated (Kaisse and Shaw 1985). Similar arguments have been made for French (Rilland 1986), Lushootseed (Urbancyzk 1996) and Trique (DiCarnio p.c.). The only example of resyllabification (1), which seems more “intuitive” to us, found so far is Latvian.
Any papers relevant to this issue are welcome. Any intuition-based data are welcome too, if you speak/study a language with vowel deletion.
Of course, in the languages that have been claimed to have the pattern in (2), there are arguments that vowels are not completed deleted—they can be either (heavily) reduced or devoiced. That sort of counterargument is welcome as well. Evidence for syllabification may be hard to come by in some languages, and people may just assume (1) or (2). That is fine and we would like to know those languages, although more explicit evidence would be welcome; we are interested in what kind of evidence has been used for syllabification in this sort of situation.
All the best,
Shigeto Kawahara (Keio University)
This paper just appeared in PlosOne: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0185534.
“Men ask more questions than women at a scientific conference”
Amy Hinsley, William J. Sutherland, Alison Johnston
Gender inequity in science and academia, especially in senior positions, is a recognised problem. The reasons are poorly understood, but include the persistence of historical gender ratios, discrimination and other factors, including gender-based behavioural differences. We studied participation in a professional context by observing question-asking behaviour at a large international conference with a clear equality code of conduct that prohibited any form of discrimination. Accounting for audience gender ratio, male attendees asked 1.8 questions for each question asked by a female attendee. Amongst only younger researchers, male attendees also asked 1.8 questions per female question, suggesting the pattern cannot be attributed to the temporary problem of demographic inertia. We link our findings to the ‘chilly’ climate for women in STEM, including wider experiences of discrimination likely encountered by women throughout their education and careers. We call for a broader and coordinated approach to understanding and addressing the barriers to women and other under-represented groups. We encourage the scientific community to recognise the context in which these gender differences occur, and evaluate and develop methods to support full participation from all attendees.
The most recent Phonolist post on this subject: https://blogs.umass.edu/phonolist/2017/10/08/question-discussion/
Facebook and Twitter seem to be the most popular platforms for academic exchange, but they have obvious problems and limitations. Do we need an open alternative to Research Gate? Check out https://www.scholarlyhub.org. Would anyone use this? We’d have to have a community there to make it worthwhile.
Their latest advisory board member Subhashish Panigrahi is an interesting guy:
“I have been working with 63 different tribes from the Indian state of Odisha speaking various diverse languages.” From “Rising Voices: Indigenous language Digital Activism”, available here: http://meson.press/wp-content/uploads/2015/08/9783957960511-Digital-Activism-Asia-Reader.pdf …
From Claire Moore-Cantwell
Here are some data gathered by Joe Pater and Yining Nie, about the questions asked at AMP 2017, at NYU. Following last year’s report, participants are broken down by gender, as well as whether they are a student or not. Unlike last year, the student/non-student cutoff is simply whether the person has a PhD or not, so post-docs and early-career linguists in visiting positions are counted as non-students. Even though it’s inaccurate, I’m using the label ‘Professors’ to cover all non-students.
Please note that the data presented below do not include data on race representation at AMP. Yining actually did collect race data on presenters and question askers (thank you!). Out of 86 questions and 19 talks, only 4 came from non-white linguists. If this extreme imbalance bothers you (and it should) I encourage you to listen to conversations that are happening at your university and others, and to consider how you can support the efforts of anti-racist activism in academia. As a start, I suggest Dr. Marcia Chatelain’s recent article in the Chronicle of Higher Ed on the topic.
138 people were registered at AMP, including 54 faculty and post-docs, 78 graduate students, and 6 undergrads. These participants were 62 female, 67 male, and 7 no response. Thanks to Gillian Gallagher for these numbers!
The demographics of those giving presentations (‘Talkers’) skewed slightly more male than female, and slightly towards professors. This graph includes three invited speakers, two women and one man.
Questions came overwhelmingly from faculty this year, although they were nearly exactly balanced by gender. (Compare to last year where the balance between faculty and students was more even, but the questions skewed much more male) Also, there were quite a bit fewer questions overall this year than last year, though the number of talks was the same: 86 this year vs, 119 last year.
Interestingly, although the actual questions were balanced by gender, hands raised for questions were not balanced (thanks to Kie Zuraw for suggesting we do these counts). Hand-raising data is somewhat difficult to collect (thank you Joe!), so we don’t have details beyond just overall gender counts for this one.
Often when I talk about question-asking stats like those of last year’s AMP, where more men asked questions than women, people assume that this is happening because women aren’t getting called on as much as men. This data shows that at least for this conference, the opposite is true – 51 raised hands from women were counted, and 43 questions from women were counted total. This means that nearly every woman who raised her hand got to ask her question. On the other hand, 126 men raised their hands total, and only 43 men asked questions. This graph also shows that women are far more conservative as a group than men in raising their hands for questions. Perhaps women only raise their hand when they are fairly certain they will be called on? (they are in the front, it is the first question, there are no other hands raised, etc.) What these data definitely show is that solving the gender gap in many conference question periods is not going to be as simple as preferentially calling on women.
For those who would like to see the data broken down a little more by individual talks, here is a graph showing the breakdown of questioner types for each talk. The gender and career status of the talkers is noted along the x axis.
Order of questions was also recorded. In all of the following graphs, numbers within a node indicate the probability that questioning started with a member of that category. Probabilities of all arrows (transitions) out of a single unit sum to 1, and are based on counts of bigrams over all question periods that begin with that category. First of all, the transitional probabilities between male and female question askers is basically at chance this year:
After a member of either gender asks a question, questioning is about equally likely to transition to another member of the same gender, or to transition to the other gender. Men are more likely to begin the question period though.
This year, there were rather few questions from students, and this is reflected in the transitional probabilities observed between students and professors:
Professors were more likely to begin the questioning than students, and no matter who asked a question, questioning was far more likely to transition to a professor.
Here is a graph of the ‘flow’ of questioning between all four categories:
And finally, the same graph, but including questions from both this year and last year (205 questions total).
The raw data on which this year’s graphs are based, as well as data from other conferences, are available at the google spreadsheet linked below:
I’d like to encourage anyone who is collecting data of this type, or who wants to, to also contribute to the spreadsheet. You can request editing access from within the sheet. I’m hoping that having (relatively) standardized data from many different conferences will start to make it easier to compare different conferences, different conference setups, and maybe different fields/subfields. Eventually it would be great to start understanding what kinds of conference configurations encourage more question asking from women, minorities, and students.
From Juliet Stanton
Donca Steriade and I are curious about some aspects of flapping in English. To respond, you can either comment on this post or email me directly at juliets [at] mit [dot] edu. Please also indicate whether or not you are a native speaker of English, and if so, what your native dialect is.
1. Do you flap in words like sanity, societal, parietal, skeletal, palatal?
2. Do you flap in words like normative, ablative, associative, additive?
3. Do you flap in words like completive, locomotive, emotive?
(We welcome any additional comments you might have on the structure of these words.)
From Kie Zuraw: Phonolist has been a serious forum, but in the spirit of the old days of Phonoloblog, where miscellaneous phonological observations shared space with weightier matters, here is the first–and most frivolous–in a short series of posts from me.
The other day, I started typing Wagner lyrics, in IPA, into Microsoft Word for a phonetics class exercise.* Imagine how impressed I was when this alert popped up:
Just a lucky guess then, but it does make me wonder what the MS language detection software is doing.
From Claire Moore-Cantwell
Below are some graphs of gender data collected at AMP 2016, by Jesse Zymet and Eric Bakovic. Note that unlike previous reports of this type, participants are broken down by gender, but also by whether they are a student or a professor. Post-docs and early-career linguists in visiting positions were counted in the ‘student’ category, although this choice does not change the numbers much.
To start off, I would like to point out that these graphs do not contain information about racial minorities, because phonology (and linguistics in general) is still overwhelmingly white. The fact that there are not enough racial minorities at our conferences to even include in the tallies is a serious problem with our field that should not be forgotten about.
Preregistered participants at AMP 2016, which were about 90% of total registrants, consisted of 59 male and 56 female participants. 72 participants were students and 43 were faculty. (Thanks to Karen Jesney for these numbers!)
Overall, AMP 2016 had a fairly balanced set of presenters (“Talkers”), though male presenters were slightly in the majority. These numbers include the invited speakers: two male and one female.
Questions came more from faculty than from students by a small margin, and more from men than from women by a much larger margin.
Note that the difference between men and women was greater among students than among faculty (a difference of 18% vs. 9%), and the difference between students and faculty was greater among women than among men (10% vs. 4%).
Below is the number of questions from each demographic, broken down by talk. Talks are arranged from left to right in order of occurrence in the conference, and speaker characteristics are notated below each bar.
Note that this graph does not contain information about the order of questions asked for each talk. That information is represented in the graphs below. First, let’s consider the transitional probabilities between faculty and students:
‘S’ indicated students, and ‘P’ indicated faculty (‘professors’). The probability inside each circle indicates how likely that category is to begin the questioning. Each arrow indicates the transition probability (or forward bigram probability) between two categories of questioner. For example, after a student asked a question, there was a 53% chance that a faculty member would ask a question and a 47% chance that another student would ask a question.
Interestingly, students are much more likely than faculty to begin a question period, but faculty end up asking more questions. This graph yields some insight into why: After a student asks a question, students and faculty members are about equally likely to ask the next question. However, after a faculty member asks a question, there is only a 29% chance that a student will then ask a question. Once the floor transitions to a faculty member, it tends to stay with other faculty members rather than transitioning back to students.
There are a couple of possible reasons for this dynamic – first, it might be explicitly encouraged via actual exhortations for students to talk first and faculty to hold back. I don’t remember this actually happening at AMP, but a second possibility is that the group implicitly follows this ‘rule’ simply because it’s common in linguistics, and the group generally believes in it. Yet another possibility is that students ask the first question more often just because there are more students in the audience.
Below is a similar graph illustrating question asking dynamics among the two genders for which data was gathered.
Men and women are nearly equally likely to ask the first question, with only a slight advantage for men. However, after a woman asks a question the next questioner is more likely to be a man than another woman. On the other hand, after a man asks a question, the probability that a woman will ask the next question is quite low.
Finally, here is a graph illustrating the ‘flow’ of questioning between all four groups.
The same general flow of floor time towards faculty and towards men that the previous two graphs showed can be seen in this graph as well. Additionally, this graph reveals that the transition of floor time between female students and female faculty looks somehwat different than the transition of floor time between male students and faculty. For both genders, after a student asks a question the probability that the next question comes from a student of the same gender or a faculty member of the same gender is about equal. After a female student asks a question, female faculty members and other female students are about equally likely to ask the next question, and the same is true for male audience members. However, after a faculty member asks a question, the probability that the next question will be from a male student is much higher (29% after male faculty, and 14% after female faculty) than the probability that a female student will ask a question (0% after female faculty, and a mere 8% after male faculty).
From Shigeto Kawahara
My students and I wrote up a paper on a few sound symbolic patterns in Japanese pokemon names, where we show that mora counts and voiced obstruents affect pokemon characters’ weight, size, and strength. The paper can be downloaded at http://ling.auf.net/lingbuzz/003196 (see also the separate Phonolist post for the abstract).
A natural follow-up question is, what about English names? I did a quick and dirty analysis addressing the same question about voiced obstruents using English pokemon names, based on orthography, which is attached to this email. Although the correlations are there, they are much weaker. But there’s an obvious flaw in this analysis; since the numbers of voiced obstruents are counted based on orthography, for example, “pidgotto” is considered to have two voiced obstruents, one for “d” and one for “g”.
Some other linguists have an intuition that in English, vowels are crucial. But obviously, automatically extracting phonetic vowel quality from orthography is not an easy task.
The bottom line: I am looking for interested researchers (or students) for collaboration who would convert the English names to some kind of transcription so that we can do a similar analysis in English. Of course if somebody can do a different language other than English, that’d be much welcome as well.
Also I’d love to have an open forum for discussion of this general project.