DELPH-IN

EMB:

FCB, SSH, and I have students working on projects, and we want those projects to be good for their career; and have a clear path to completion.
I feel that just building a grammar without a research question doesn’t make a good PhD.
OTOH, if the question is “what do you have to do build a grammar for X” and the answer is a bunch of small analyses, that’s also not satisfying.
Joshua, Glenn, and David are working on grammars; SSH did a library; but Joshua’s Negation library was not well-received by one committee member.

Luis:

We have people saying, why don’t you parse 100%? where’s the value

FCB:

No, they’re asking “why don’t you handle discourse?” or some other phenomenon
We want the topic to be something that helps the student get a job later.

EMB:

We don’t tell students what to do, but we tell them “this won’t help you in the job market”; maybe the student doesn’t mind
So what we do is choose 2 or 3 central phenomena, and build them in a CLIMB-based approach (test different analyses); this focuses the dissertation on one thing, but you can talk about how some analysis affects overall coverage, etc.
This isn’t the only way, but it’s what we’ve locally decided works for grammar development

Glenn:

Is it artificial to force students to explore alternate analyses if they already know what they want?

EMB:

It is somewhat artificial, but often we don’t know what is the best approach.
Wambaya is one example where I asked “how could I do this differently?”; while working on alternate analyses, I discovered bugs, but instead of introducing those bugs into both analyses, I forked the grammar in order to test it. Antske took the overall process of testing multiple analyses and formalized it somewhat, so that choosing alternatives is not bound by the order of development.

Luis:

We are concered with the long-term health of the grammars, so if we say “we want a Chinese grammar”, we might ask students to tackle certain phenomema, but they might disagree with previous analyses for other parts of the grammar…

EMB:

Another kind of thesis is a “grammar development methodology” thesis.

Luis:

How can one develop a grammar for one language without going for coverage?

EMB:

You can go for coverage, but you need to focus the dissertation on one or several particular phenomena.

Glenn:

Is it the case that these student grammars are left for dead after the defense?

Luis:

I think it’s rare for grammars to be developed after the initial work.

Glenn:

Would a dissertation focused on one phenomena rather than overall grammar coverage and health make that problem worse?

EMB:

It seems to me you’d like students to come along and pick up different phenomena. Even Dan says that certain parts of the ERG are placeholder analyses. Why do you assume students can’t come in a pick up a different part of the grammar.

Luis:

My point is, I see lots of grammars without very active development. My guess is that it’s hard to build a PhD on top of an existing grammar.

EMB:

I haven’t had students come to me and say “I want to do a Japanese phenomenon; can I work on Jacy?”. I find that student come in with different language interests, so they work on their own grammars.

ZZ:

I’ve recently read Joanna(?)’s thesis; she focused on the BA analysis (Mandarin). Of 300 pages in the thesis, 200 described the phenomenon, only 60 described HPSG. From a linguistic point of view, that is a very respectable body of work. But is it generally useful for the Chinese grammar? Of course she needs more than BA. She associated it well with other parts of the grammar, but those parts need to be implemented also.

EMB:

I’d like to think that if she did all that work, it would be clear from the thesis.

Glenn:

You need to anticipate that from the beginning. E.g., I need to do some work on tools to support the development I want to do.

FCB:

I’m not sure I agree with EMB that the Question is the only way. There are multiple kinds of PhDs. Building something of lasting utility, etc.

EMB:

Where’s the narrative in that dissertation worth reading?

FCB:

Few dissertations are worth reading…
For ZZ we had lots of methodological things to solve; for David (Moeljadi) we considered MT and integrating KBBI, etc.
For the descriptive theses I advise, I haven’t seen the Question approach. There’s a template of doing (a) phonology, (b) morphology, (c…), and some students focus on one of those more than others
We are making a new kind of linguistic description that wasn’t possible before.

Luis:

There are subfields; you can look at the grammar as more linguisticky or more computational. If you approach it as “this looks like its fun to work on”, when does it end?

FCB:

4 years

Luis:

Where’s the sense of accomplishment if after months of hard work you parse 60% of the corpus?

FCB:

(some book says) “there’s only one kind of thesis, it must focus on (some narrow subfield of linguistics), must be framed in this way, …, otherwise it’s nonsense.”
not everyone shares these views

Glenn:

in addition to the thesis itself, our grammars are a useful artifact of the process. Do we get credit for that?

EMB:

no
some people do descriptive work and get grammaticality judgments and don’t share that data; it’s incomplete
when you do theoretical syntax, it’s pages of dense argumentation, and a literature review where you’re accountable to every point, and what we’re doing is different in that regard

ZZ:

we can phrase it as “what are you contributing?”
if it’s “we are advancing understanding of BA”, that’s a good thesis
if it’s “we have a grammar that now covers X% of the corpus”, people might say “well others have done that (e.g. in LFG), so what’s new here?”

Glenn:

None of us wonders if it’s possible to complete a grammar (e.g. to 100%), so that’s not a goal

FCB:

Many theories of language are based on a small subset of languages, so working on lesser known languages could have value in advancing general understanding
My thesis was a problem-solving thing. “How can we do better than the state-of-the-art”? I said “if we introduce these phenomena (countability, coreference, etc), we can improve results..”

Glenn:

Our work falls into “engineering”, and so it’s not necessarily about doing something novel so much as about building something

EMB:

Science is about finding truth and engineering is about finding solutions; either way you should generate new knowledge.

FCB:

you need to generalize your findings

EMB:

important Q: you claim to produce new knowledge; who is that interesting to and why? Frame it so the value is obvious.
if the knowledge we generate is completely locked in the artifact (grammar or system), we have not done our job.

FCB:

not everyone considers all ways (of doing a thesis) are as valid as others; consider who’s on your committee, your examineers, and your hirers (potential employer)

EMB:

and who you want it to be useful to later

Olga:

and you need something that you will be comfortable with

EMB:

what happens to the people who do pure-descriptive theses?

FCB:

they go back to their home country and teach English, or linguistics
they aren’t necessarily the theses we’re proudest of, but many of these students come from home universities without a strong research program
but sometimes they uncover things that are typologically interesting
everyone wants to redo the orthography first, though

Olga:

the problem of wanting to redo things is a very old problem, even in CS, it ends up costing more to rebuild rather than reuse

EMB:

you need to spend your time on what’s your novel contribution, so don’t waste time if redoing things doesn’t advance your overall progress
redoing orthography because the existing one doesn’t represent a phonological distinction is important; but maybe it’s not what you want to focus on

FCB:

there are some changes where you specialize
some people aren’t satisfied until they’ve rewritten some piece of code into another (and another) programming language…
there’s disagreement on what a good PhD thesis is regarding novelty; some think that the student must do *everything* on their own; others are fine if you’re in a team of 50 people on some bigger problem and you’re in charge of some small part of it.

David:

I like grammar engineering for my thesis because my background is linguistics, and I get to play with various phenomena in my language, but also because the grammar produces results. I get to see trees and semantics from the grammar I built.

Glenn:

so you think you couldn’t do all the linguistic work you want to do without the computational approach?

David:

well I like that it feels like language documentation but it works

FCB:

60% coverage was a stretch goal, but for a PhD you don’t have to do what you started out to do. You need something at the end, but it could be very different from the original goal.
btw, with robustness measures, we’re probably close to 60%, so we can hit that number of we really wanted to

EMB:

yes, these things are negotiable, but must be negotiated. The student shouldn’t go off to their tower for 3 years without communication with their committee.
Martin Kay is an extreme example for coming back after many years with some body of work and getting a PhD
generally students need lots of updates and adjustments along the course

Luis:

Essentially the advisor says “it’s your phd, I’m here, but good luck.”
This is nice when doing something theoretical, because you need the space to develop your theories (e.g. when they disagree with your advisors’ ideas)

EMB:

Joshua and I have longstanding disagreements about how to represent things in MRS
I say “fine, but you have to say why you’re doing things differently; you can’t just describe the difference”

Luis:

You need to defer to your advisors; your advisor is not a peer. It’s in your interest to maintain a healthy relationship.

FCB:

There are examples of disfunction between departments because of strong differences of opinion…

EMB:

Not all advisors understand their role and responsibility; there’s not a lot of training for how to be an advisor

FCB:

let’s return to “how do we design PhDs that contribute to the field as a whole?”
it ties in with “how can we build grammars?”

EMB:

there’s the question of “where can we publish?”
the ERG is an example of something that’s done a lot but not written about much of it
there are pieces that would not constitute a book, but would be great articles, and this is generally true for other grammars
there’s a scale of linguistic understanding that we can only do with a computer, and we shouldn’t be shy about that

David:

but if you write up an HPSG analysis of something, that’s only of interest to HPSG people

Glenn:

but maybe our approach is required for the work we do, because of that scale

EMB:

no, and it’s not worth putting that point in a paper, because then you have to back it up

…

David:

how many languages have enough resources to sustain grammar development?

FCB:

with a competent native speaker, all of them

EMB:

but then you’re doing the fieldwork and the grammar development, and that’s much harder, as David can attest

Luis:

I’ve been in sad conversations (e.g. at QTLeap) about grammar maintenance; e.g. I’m not sure it’s ok to spin off Jacy for fear it would upset Jacy’s maintainers, etc.

FCB & EMB:

we wouldn’t be offended

Last update: 2017-01-06 by MichaelGoodman [edit]