Speech genre and context efects in
sentence processing
What do you expect when you’re expecting pop songs?
Lauren Squires
The Ohio State University
[email protected]
http://u.osu.edu/squires.41
Sociolinguistic Variation and Language Processing
Virginia Tech
April 2, 2016
BACKGROUND
Grammatical variability and sentence processing

Less familiar grammatical structures cause processing
difculty
 People quickly learn to comprehend novel grammatical forms and
may even be able to generalize to diferent forms
▪
Kaschak & Glenberg 2004; Kaschak 2006
 Alternatively, maybe people just learn to tolerate previously bad-
seeming structures
▪
Luka & Barsalou 2005; Francom 2009; Luka & Choi 2012; Boland et al. 2015
BACKGROUND
Social information and sentence processing

Social cues infuence syntactic and semantic processing
 talker specifcity
▪
Kamide 2012; Squires 2014b*; Yildirim et al. 2016
 age+sex
▪
van Berkum et al. 2008; Tesink et al. 2009
 race/ethnicity
▪
Staum Casasanto 2009*; Seifeldin et al. 2015*
 native/nonnativeness
▪
Hanulíková et al. 2012
*investigate “canonical” sociolinguistic variation/variants
BACKGROUND
Grammatical variability and social information
in sentence processing

Seifeldin et al. 2015: Processing variable copula deletion
(feature of African American English)
 ERP during listening task
 Grammatical manipulation:
▪
standard: ...he’s going...
▪
absent copula: ... he going...
 Speaker race/ethnicity manipulation:
▪
White, African American, Indian
 P600 (syntactic anomaly) efect for copula-absent sentences for
White speaker condition, but not African American or Indian
BACKGROUND
Social information as “expectation-loosener”

Social information can facilitate processing less-familiar
structures (mitigate processing costs)
 “...listeners found copula deletion ungrammatical when listening to a
Standard speaker, but not when listening to a speaker of a non-standard
variety of English, regardless of whether the specifc variety they hear is
characterized by rules allowing for this construction. ... listeners do not apply
dialect-specifc knowledge on-line when processing the syntax of a nonstandard variety of their native language; rather, listeners loosen their
expectations for standard syntax.“ (Seifeldin et al. 2015; see also Hanulíková et al. 2012)

Grammatical expectation is manipulable via contextual cues
 The relevant contextual cues should be known to bear some
experiential relationship to grammatical structures
▪
though this may be fairly abstract and nonspecifc knowledge
▪
on prediction / expectation in language processing: Kuperberg & Jaeger 2016
(Language, Cognition, & Neuroscience 31(1) special issue)
THE STUDY
Research Questions

What social factors cause expectations of
grammatical form to shift?

What is the relationship between linguistic and
social factors on expectations of grammatical form?
THE STUDY (A BACKSTORY)
Social information  speech genre  song lyrics

“Where can I fnd (lots of) real sentences containing nonstandard
grammatical features? that a researcher didn’t invent? and that have
other properties useful for experimental endeavors?”
 Corpora document more formal registers where nonstandard variants
are infrequent
 Grammatical variables are typically less frequent than phonological
ones anyway

Song lyrics
 a speech genre in which nonstandard grammatical variants occur frequently
and may even enjoy a kind of covert prestige
▪
Eberhardt & Freeman 2015; Beal 2009; Trudgill 1983
THE STUDY
Social information  speech genre  song lyrics

Song lyrics as speech genre
 In Anglophone popular (commercial) music, part of the context-form link
constituting the genre is the licensing of nonstandard phonological and
(morpho)syntactic variants
 Pop music is a circumscribed speech genre treated “specially” in terms of
standard language ideologies
▪
▪
Even “super standard” English speakers often listen to, tolerate, and even sing along
with song lyrics containing nonstandard grammar
One might argue that nonstandard features are enregistered as part of the genre
▪

e.g., Dr. Squires’ Double Negation Medley
Speech genre as social information that could shift grammatical
expectation
 If people have knowledge that nonstandard grammar is part of the genre,
then expecting song lyrics should “loosen” the expectation for standardness
THE STUDY
Experiment design

Target structure: NPSG+don’t (invariant don’t)
 Common across dialects of English worldwide; common in song lyrics
 For “standard” speakers, has processing cost and triggers evaluation of lower
social status (Squires 2013, 2014a, 2014b, 2014c, forthcoming)

Self-paced reading (moving window)
 Participant presses a button when ready for the next word
 Paradigm software; Dell computers; Cedrus response pads; English
Linguistics Lab

152 sentences total
 40 target – 20 nonstandard (NPSG+don’t)
 8 practice, 104 fller – 32 nonstandard fllers across 8 nonstandardisms

40 comprehension questions following fller sentences
 “Which of these words did NOT appear in the previous sentence?”
[word1] [word2]
THE STUDY
Experiment design

Within-subjects manipulation: Standardness
 standard: NPSG+doesn’t
 nonstandard: NPSG+don’t

Between-subjects manipulation: Context
 Context participants told that sentences come from song lyrics
 NoContext participants told nothing about the sentences
THE STUDY
Context manipulation
1
In this experiment, you will be silently reading a series of sentences.
The sentences come from song lyrics from a number of genres.
These genres include pop, country, rap, rock, folk, hip-hop, alternative, and R and B.
Each sentence will appear one word at a time.
You will use the middle green button on the response pad to advance through the
2 lyrics.
Once a word appears, read it as quickly as possible, then press the green button to
move on to the next word.
As you go through the experiment, you will be asked questions about some of the
3 sentences.
Use the red (left) and blue (right) buttons on response pad to answer the questions.
Be sure to pay attention to the words in the lyrics!
4 The following sentences are for practice.
They do not come from song lyrics.
THE STUDY
Hypotheses
Standardness efect: slower reading times for don’t than
doesn’t
1.

nonstandard slower than standard
Context efect: efect of standardness smaller for Context
than NoContext participants
1.

context facilitates expectation shift away from “standard” grammar
THE STUDY
Song lyrics as stimuli: special considerations

Content
 Avoided extremely well-known, iconic lyrics; tried to take lines from
verses rather than choruses; tried to use older songs from older artists
 Tried to avoid explicit references that could be generic giveaways
●

whiskey, guns, trucks, dance clubs, drugs, sex, music, heartbreak...
Form





Target sentences always “DET N don’t/doesn’t V/ADV...”
Edited for length (4-9 words)
Edited to remove other nonstandardisms
Edited for conciseness
Nonstandardism in fller sentences varied across word position
THE STUDY
Stimuli: examples
nonstandard
standard


The boss don't care about the dress code
This industry doesn't come with benefts
gonna
FILLERS
tryna
gotta
wanna
you was
they was
demonstrative them
there’s+PL
no nonstandardism









Too much cofee's gonna shatter my nerves
They're tryna break the marriage up
This has gotta be the good life
I wanna fnd that perfect moment
First you was in the sky
This is the one they was waiting on
Them kind of monkeys can't swing
There's too many rivers to cross
The ferris wheel carried me away
TARGETS
THE STUDY
Participants

Linguistics subject pool

Extra credit for English classes

97 total
 48 NoContext
 49 Context
THE STUDY
Analysis

Practice trials removed

Outliers removed: under 100 or over 2000 ms
 <.25% of observations

Residual reading times
 create linear regression model for raw RT, including all target and fller
sentences
▪
▪
fxed efects of word length in characters and word position in sentence
random intercept term for subject
 use residuals of that model as dependent variable (Jaeger 2008)

Mixed-efects linear regression with lmer() for each word region
 random intercept terms for subjects and items
 stepwise analysis for fxed efects of trial order, standardness, context, and
their interactions
RESULTS
Targets
Standardness
don’t/doesn’t
RESULTS
Targets
Trial Order : Standardness
Target Sentences: Residual Reading Times by Trial Order and Standardness
Residual Reading Time (ms)
40
grammarcond
0
nonstandard
standard
40
50
100
Experimental Trial
150
RESULTS
Targets
Word-by-Word Results
Word Region
Signifcant Fixed Efects
Estimate
Level
word 2
trial order
-0.69003
<.001
word 3
(don’t/doesn’t)
trial order
standardness
standardness : context
trial order : standardness : context
-0.80372
-27.81207
45.13892
-0.36425
<.001
<.1
<.01
<.05
word 4
trial order
context
trial order : context
-0.83046
16.59428
-0.17022
<.001
<.05
<.05
word 5
trial order
-0.77095
<.001
word 6
trial order
-0.91390
<.001
•
•
Fixed-efect factors retained from stepwise analysis
All models include random intercept terms for subjects and items
RESULTS
don’t/doesn’t

Trial Order (<.001)
 Later trials faster (<.001)

Standardness (<.1)
 Standard doesn’t faster than nonstandard don’t

Standardness : Context (<.01)
 Standardness efect larger for Context group than NoContext group
Targets
Standardness : Context
RESULTS
don’t/doesn’t

Trial Order (<.001)
 Later trials faster (<.001)

Standardness (<.1)
 Standard doesn’t faster than nonstandard don’t

Standardness : Context (<.01)
 Standardness efect larger for Context group than NoContext group

Standardness : Context : Trial Order (<.05)
 Standardness : Trial Order interaction only relevant for NoContext group
▪
For NoContext group, standardness efect increases over course of experiment
RESULTS
don’t/doesn’t
Trial Order : Standardness : Context
DON'T/DOESN'T: Residual Reading Times by Trial Order and Standardness across Context Group
context
nocontext
Residual Reading Time (ms)
50
0
grammarcond
nonstandard
standard
50
100
50
100
150
Experimental Trial
50
100
150
DISCUSSION
Discussion

H1 confrmed (in interaction)
 Nonstandard read more slowly than standard

H2 not confrmed
 Standardness efect larger for Context group than NoContext group
 Three-way interaction: context, standardness, trial
▪
For NoContext group, standardness efect shows up later
DISCUSSION
Expectation v. Attention v. Efciency

Unexpected (!) fndings
 Standardness efect greater for Context group
▪
Maybe context information heightened attention—people paying more
attention to individual words, therefore more noticing of nonstandardism
▪
▪
BUT: in raw RTs, Context group faster overall than NoContext group
AND: no diference in comprehension question accuracy
 Standardness efect develops later for NoContext group
▪
Maybe size efect is due to a stability efect: context information leads to
more efcient/consistent processing behavior that makes the grammatical
efect easier to detect
▪
Additional trials at doing the task make both groups faster, but for those with no
context going into it, additional trials also stabilize grammatical efect
DISCUSSION
Insight from post-experiment comments
“Did you notice anything interesting about the sentences you just read? Please describe.”
Representative comments: NO CONTEXT group

There were grammatical mistakes in a lot of them that caught my eye. Also a lot
of them sounded like quotes from something (i.e. song lyrics).

Sometimes the grammar wasn't 100% correct (ex. using "them" instead of "the"
as a defnite article) and some of the sentences just didn't make much sense (like
the one about the clown honking someone's nose or something? that was weird).

They had no correlation, in my opinion, at frst, though some of the sentences
did go together.I also noticed the use of reptition of soe words in the sentences.
It also seemed as if a few went together, as they rhymed.

they were gramatically incorrect. In my mind, I read them with a southern accent
because the writting looked southern.
DISCUSSION
Insight from post-experiment comments
“Did you notice anything interesting about the sentences you just read? Please describe.”
Representative comments: CONTEXT group

They weren't all syntactically well-formed. Some of them didn't make sense in
terms of syntax. And some of them I knew which songs they were lyrics to.

I noticed that the sectences that I just read were possibly hip hop or country
lyrics. They weren't what most people consider "proper English."

I thought I would recognize more of the lyrics but I really only recognized two of
them.

The only thing that stood out to me was the word choice in the sentences. It is
clear that there was some deviation from Standard American English. But that is
to be expected in song lyrics.
DISCUSSION
Insight from post-experiment comments

Both groups noticed nonstandard grammar and thought the content
was “odd”
 Some participants characterized all the sentences as having “bad grammar,”
despite the fact that more than half of them were standard
 For many Context participants, the nonstandardness and content oddities
were both anchored to song lyrics as a speech genre, or to specifc sub-genres
of music
 Several NoContext participants recognized lyrics

Both groups sought to make coherence out of the sentences
 NoContext participants tried to identify a genre
 Context participants tried to recognize familiar sentences from memory
DISCUSSION
Expectation: Form v. Content

Context information did not shift expectation of grammar itself
 although participants did articulate a link between the genre and the
grammar

Context information did shift expectations of:
 what kind of content they might encounter
 what the relationship between the sentences they would read should be
 something about how to read the sentences

Just as in social life, participants want to know “What’s going on here?”
 With context information, participants had an idea right away, established a
steady baseline—and read for recognition of content
 Without context information, participants spent the frst part of the
experiment fguring it out—and read for understanding the task
iiiiiiiiiii want to thank you






Jamie Smith
OSU LOC (Hope Dawson, Kiwako Ito)
Roger Cherry
Bruno Tagliaferri
Julie Boland and Robin Queen
SVALP
[email protected]
References
Beal, Joan C. (2009) "’You’re not from New York City, you’re from Rotherham’: Dialect and identity in British indie music." Journal of English Linguistics
37.3: 223-240.
Boland, J., de los Santos, G., Carranza, J., & Kaschak, M. (2015) Self-paced reading time as a measure of learning novel constructions. 28th Annual CUNY
Conference on Sentence Processing. University of Southern California, March 19-21.
Eberhardt, M., & Freeman, K. (2015). ‘First things frst, I'm the realest’: Linguistic appropriation, white privilege, and the hip-hop persona of Iggy Azalea .
Journal of Sociolinguistics, 19(3), 303-327.
Francom, J. C. (2009). Experimental syntax: Exploring the efect of repeated exposure to anomalous syntactic structure - evidence from rating and reading
tasks. Tucson: University of Arizona Ph.D. Dissertation.
Hanulíková, A., Van Alphen, P.M., Van Goch, M.M., & Weber, A. (2012). When one person’s mistake is another’s standard usage: The efect of foreign
accent on syntactic processing. Journal of Cognitive Neuroscience, 24(4), 878-887.
Kamide, Y. (2012). Learning individual talkers’ structural preferences. Cognition, 124, 66-71.
Kaschak, Michael P. 2006. What this construction needs is generalized. Memory & Cognition 34(2). 368-379.
Kaschak, Michael P. & Arthur M. Glenberg. 2004. This construction needs learned. Journal of Experimental Psychology: General 133(3). 450-467.
Kuperberg, G. R., & Jaeger, T. F. (2016). What do we mean by prediction in language comprehension? Language, Cognition and Neuroscience, 31(1), 3259.
Luka, B. J., & Barsalou, L. W. (2005). Structural facilitation: Mere exposure efects for grammatical acceptability as evidence for syntactic priming in
comprehension. Journal of Memory and Language, 52(3), 436-459.
Luka, B.J., & Choi, H. (2012). Dynamic grammar in adults: Incidental learning of natural syntactic structures extends over 48 h. Journal of Memory &
Language, 66, 345-360.
Staum Casasanto, L. (2009). Experimental investigations of sociolinguistic knowledge. Stanford, CA: Stanford University Ph.D. Dissertation.
Seifeldin, M., Cantor, M, Boland, J., & Brennan, J. (2015) They dropping copulas: Salient cues in the integration of speaker identity and syntax. 28th
Annual CUNY Conference on Sentence Processing. University of Southern California, March 19-21.
Squires, L. (2013). It don't go both ways: Limited bidirectionality in sociolinguistic perception. Journal of Sociolinguistics 17(2), 200-237.
Squires, L. (2014a). Knowledge, processing, evaluation: Testing the sociolinguistic perception of English subject-verb agreement variation. Journal of
English Linguistics 42(2): 144-172.
Squires, L. (2014b). Talker specifcity and the perception of grammatical variation. Language, Cognition and Neuroscience 29(7): 856-876.
Squires, L. (2014c) Social diferences in the processing of grammatical variation. Penn Working Papers in Linguistics 20(2): Article 20. (Selected Papers
from NWAV42) http://repository.upenn.edu/pwpl/vol20/iss2/20/
Tesink, C. M., Petersson, K. M., Van Berkum, J. J., Van den Brink, D., Buitelaar, J. K., & Hagoort, P. (2009). Unifcation of speaker and meaning in
language comprehension: an fMRI study. Journal of Cognitive Neuroscience, 21(11), 2085-2099.
Trudgill, P. 1983. Acts of conficting identity: The sociolinguistics of British pop-song performance. In Peter Trudgill (ed.), On dialect: Social and
geographical perspectives, 141-160. Oxford, UK: Blackwell.
Yildirim, I., Degen, J., Tanenhaus, M. K., & Jaeger, T. F. (2016). Talker-specifcity and adaptation in quantifer interpretation. Journal of Memory and
Language, 87, 128-143.
Van Berkum, J. J. A., Brink, D. van den, Tesink, C. M. J. Y., Kos, M., & Hagoort, P. (2008). The neural integration of speaker and message. Journal of
Cognitive Neuroscience, 20(4), 580-591.