9.59/24.905 Lab in Psycholinguistics
Spring 2013
Instructors: Edward A Gibson, Peter Graff, Lydia E Volaitis
TA: Kyle Adam Mahowald
Lecture: TR10.30-12 (46-1015)
Information:
In this class, students learn to design, conduct, analyze and present experiments on the structure and processing of human language, through hands-on experience. The main focus of the class is on constructing, conducting, analyzing and presenting an original and independent experimental project of publishable quality. We will also develop skills in reading and writing scientific research reports in cognitive science, including evaluating the methods section of a published paper, reading and understanding graphical displays and statistical claims about data, and evaluating theoretical claims based on experimental data.
The topics of research will all involve the structure and processing of human language. Human language is structured at many levels: sounds are structured into morphemes; morphemes are structured into words; words are structured into phrases and sentences; and sentences are structured into discourses. We will review a variety of studies of different factors influencing structure and processing and all of these levels of representation
Research Projects
The first half of the class is devoted to understanding and replicating a published experimental result. Students choose a topic on human language research from a list provided by the instructors. Students design and conduct a replication of the published experiment. Students are expected to complete the replication by spring break. In the second half of class, with guidance from the instructor and TAs, students design, implement and analyze the results of an original experiment, which extends the first project in a novel direction. Emphasis is placed on original and genuinely important extensions that produce publishable data.
Students are encouraged to construct and conduct the experiments during lab time, when a TA or instructor will be available, along with computers and other technical tools.
At the same time that students are choosing their topic for research, there are lectures presenting background on experimental research in language stucture and processing from the existing literature. Language has many inherent components, such as the words, the syntax (word order), the meaning, and the intonational contour (for auditory materials). All of these factors may vary in their frequency in the input, which may affect their complexity in understanding or production. Consequently, any experiment that investigates one aspect of human language – e.g., a particular syntactic construction within a language – must also control for all the other factors that are known to affect the kinds of dependent measures that are being investigated in the experiment. Thus no matter what topic a student may choose to investigate, s/he must know about other factors that are involved in naturalistic human language, and control for them as much as possible in their experimental designs.
Mechanical Turk
Most students will run their experiments on Mechanical Turk, a
crowd-sourcing website available through Amazon.com, where one can
get experimental participants very quickly. The department will pay
to run these experiments.
Announcements
Projects
Hi all, your project extension should be up and running on Turk by next Friday's lab (May 3). The week after that should be spent on analysis. You should feel free to send us materials and questions over the next few days.The final paper should include:
-introduction with background on the paper you are replicating as
well as related papers (you need to read and write about at least 3
other related studies from the literature)
-methods and results for your replication
-methods and results for your extension
-a discussion section
You should start on all of this soon! Final papers are due May 16, and your presentations will be May 14 and May 16.
Announced on 26 April 2013 11:28 a.m. by Kyle Adam Mahowald
This week
Hi, a couple announcements:1. At the end of class today, I said I was getting 910 rows for Pset3 #1. I'm now getting 886. There is one subject right around .9, and it differs based on the order of steps. You can do it either way, but I recommend excluding individual NA's as the last step. We won't worry about this too much, though.
2. Leon Bergen will be lecturing Thursday on semantics. There is no lab Friday.
3. Peter, Ted, and I will be away at a conference. Your problem set is due Friday. I will try to stay available via e-mail and Piazza over the next few days.
Announced on 19 March 2013 3:18 p.m. by Kyle Adam Mahowald
Papers for projects
The suggested papers for your projects are now available under Materials. The project will involve replicating and then extending the result from one of the papers. We have quick summaries of the papers below (and also in the file project_summaries.txt). You should look the papers over and give us your top 3 choices by this Wednesday, February 20. You can then meet with one of us to talk about what it might be interesting to do.Remember that Paper 1 is a review of a paper in the literature. Although it's up to you, it will probably be easiest to do that for whatever paper you choose to replicate and extend. We will talk about all this in more detail in class.
****
Goodman and Frank: Predicting pragmatic reasoning in language
games
Goodman and Frank do language game experiment showing that people
use Bayesian inference, relying heavily on context, in order to
recover a speaker's meaning.
Rohde and Ettlinger: Integration of Pragmatic and Phonetic Cues
in Spoken Word Recognition
This paper investigates how very high level factors like
pragamatics influence a very low level factor like phonetic
perception.
Scontras, Graff, and Goodman: Comparing Pluralities
Scontras et al show that people infer collective properties over
groups of object before comparing them along a single
dimension.
Schmidt et al.: How tall is tall?
Schmidt et al. present a cognitive model for how people identify an
object as "tall."
Piantadosi, Tily, Gibson: Word lengths are optimized for
efficient communication
This paper improves on Zipf's idea that a word's length is
inversely proportional to its frequency by showing that a
word's length is better correlated with its information content
in context.
Mahowald, et al.: Info/information theory
This paper the predictions of Piantadosi et al. for meaning-matched
word pairs like chimp/chimpanzee and finds that a more predictive
context encourages the shorter form.
Daland et al.: Is sonority sequencing evidence for innate
phonological knowledge?
English speakers prefer mlat to lmat even though they've never
heard either. While some researchers have attributed these
preferences to innate phonological knowledge, Daland et al present
evidence for an explanation on terms of statistical learning.
Hilpert 2008 Are morphological alternations subject to
communicative efficiency?
Hilpert 2008 shows a variety of factors that influence English
comparative formation (more proud vs. prouder). Is this alternating
subject to information theoretic considerations as well (cf
Mahowald et al. 2013).
Coetzee 2009: Consonant co-occurrence restrictions and the
nature of phonological representations.
Coetzee shows that certain long distance combinations of consonants
are dispreferred (e.g. spap is not a good word in spite of the fact
that spat and tap are). How much of their knowledge of these
restrictions have speakers inferred from the statistical properties
of their phonological lexicon?
Gibson, Bergen, and Piantadosi: The rational integration of
noisy evidence and prior semantic expectations in sentence
interpretation
Gibson et al. use a noisy channel explanation to show that
sentences are interpreted according to a rational theory of noise
in the communication process.
Bergen and Gibson: Agreement errors as rational errors in
sentence comprehension
Bergen and Gibson look at agreement errors like "The keys to
the cabinet is on the table" and find that they are consistent
with a rational theory of noise.
Announced on 17 February 2013 11:16 a.m. by Kyle Adam Mahowald
Piazza
We've added a Piazza page for the class so that you can all see relevant questions and answers about the problem set. Let me know if you have any issues signing up.Kyle
Announced on 15 February 2013 12:38 p.m. by Kyle Adam Mahowald
