In my opinion, there are two longterm goals for reading in classes.
1. Become a better writer by examining what works and what doesn’t.
Learn norms for word choice, structuring, stats, presentation, etc.
Create a list of exemplars of good writing and exemplars of bad writing.
Compare them and profit.
2. Create a mental database entry for the paper, the level of detail of which depends on its relevance.
If the paper is outside of your field, you want to at least remember (10 years later): “Author (year-ish) argues that ….” in one sentence
If the paper is related to your work, you want to at least remember (10 years later): “Author (year) argues that…, using evidence from …, and the basic structure of the argument is…” in a few sentences.
Essentially, we want to fill in these database fields:
Paper goals (as stated by author)
General argument structure
Evidence considered
Best parts of paper
Problems with paper
First step: prereading to build scaffolding
Predict the content and argument based on the title.
What is the contribution of this paper?
How does this paper fit into the existing literature?
Read the abstract and introduction carefully, again searching for the contribution and positioning the paper within existing literature.
Skim the conclusion, so you know how the paper is going to end.
Many papers follow the “say everything thrice” method, and for these, the conclusion and the introduction will contain the same content. (This is the approach advocated in Chapter 3 of Doing OT.)
Look at the road map at the end of the introduction, flip through the paper and predict how each section is going to fit together and build (or relate to) the central argument.
What is the structure of the argument?
Example 1 (Paster 2006):Phonologically conditioned allomorphy (e.g. a vs. an) is either listed in the lexicon or derived via the grammar (but not both). Here are cases that cannot be derived via the grammar (and crucially can be listed in the lexicon), therefore allomorphy must be listed in the lexicon.
Example 2 (Smith 2015):If phonologically conditioned allomorphy (e.g. a vs. an) is always listed in the lexicon, there always needs to be enough data to learn it. Here is a case where there aren’t enough data to learn it, therefore it can’t always be listed in the lexicon.
Example 3 (McCarthy et al. 2011):Assume Harmonic Serialism (a version of OT with a different GEN function). Underlyingly linked tones in Harmonic Serialism make it impossible to account for a phonological process in Kikuyu, therefore tonal links cannot be in underlying forms (at least in Harmonic Serialism). Without underlying tonal links, we can still account for some other languages that look like they need tonal links in underlying forms.
Example 4 (Smith & Pater accepted):Three generative models of free variation make different predictions about possible patterns of variation. Here are some French judgment data that cannot be modeled in one of the three models, therefore one of the other two models is preferable. Of the remaining two models, one is a little better for these data, but we’re not sure why.
Think of papers and presentations – both as producer and consumer – as presenting a narrative or telling a story like those above.
Second step: fill in details and actively engage with the paper
Check the content against the expectations established in Step 1.
When you are presented with data, sketch an analysis before reading the author’s. Then see what the differenes are.
After each example, subsection, section, and major discussion of related work, stop and ponder how it relates the the main contributions described in the abstract, introduction, and conclusion.
Take notes and keep track of:
all assumptions being made and whether they are essential to the argument
whether the author does what they say they’re going to do
Example: if an author argues for model A over model B using a dataset, they need to show that both A can account for the data and B cannot.
what’s important
what’s surprising
what’s wrong
words you don’t know
references you want to (or need to) read
see the Slack post on types of questions/comments for more ideas
Examples for an OT article:
I add Ws and Ls to all violation tableaux and check ranking arguments.
I check to see that constraints that are added later in the paper don’t break the ranking arguments earlier in the paper.
I think of how the author’s constraints relate to previous rule-based analyses.
I pause after every tableau to think of missing candidates, inputs, and constraints (see Doing OT Chapter 2).
I check constraint definitions against violations to make sure I understand how constraints are assessing violations.
I keep track of all assumptions, especially ones that aren’t the norm (either becaue there is no norm or the assumption contradicts the existing one).
Example of assumptions: privative vs. binary features, syllabification in inputs, a new definition of a faithfulness constraint
I try to figure out why the author is making a particular assumption.
Is it necessary for the argument to hold?
Is it justified by previous work that presents a good argument for the assumption?
Is it an arbitrary decision to make the framework explicit?
Possibly, as a last resort: Is there a social reason for this assumption? Does the author not realize that this is an assumption they’re making?
You want to put your notes in a central place, so you can consult them many years later after you’ve forgotten the details of the paper.
This paper is tough and I’m stuck
Read it aloud.
Skip the tough section and come back to it with more context.
Still stuck?
Talk through the paper with someone: rubber duck debug.
Get more background: read background papers (or textbooks if relevant) and ask an advisor or peer for help.