Transcript slides

Saying nothing: Frequency effects
in Dominican Spanish null subjects
Cristina Martinez Sanz (U of Ottawa)
& Gerard Van Herk (Memorial U)
The spark:
• “every aspect of language can profitably be
re-examined in light of the important
frequency effects” (Bybee 2002)
• …but most work so far is on phonology
For example…
• intervocalic /d/ (Bybee, 2002; Díaz-Campos &
Gradoville, 2011)
• coronal stop deletion (Bayley & Loudermilk,
2008; Bybee 2001, 2002; Walker 2012)
• lenition of Spanish syllable final /r/ (DíazCampos, 2005, 2006)
• syllable final /s/ (Brown, 2009; Fife-Muriel,
2009)
• Some studies find frequency effects, some
don’t
• But, very little work on frequency effects in
syntax
Today’s variable
• Null subject in Spanish, aka “subject personal
pronoun” (SPP) variation
• For example:
yo/0 te voy a hacer una historia buena
‘I’m going to tell you a good story’
A widely studied variable…
• …but not with respect to frequency
Dialectal differences
• Factor groups: stable across dialects
• Factor group rankings:
-Person is the first ranked factor group in Caribbean
dialects, whereas it is overriden by discourserelated constraints (switch reference) in nonCaribbean dialects (Otheguy et al. 2007, Orozco &
Guy 2008, Martínez Sanz 2011).
Frequency effects
• Erker & Guy (2012) look at Mexican and
Dominican Spanish speakers in New York City
• 12 informants, 4916 tokens
• “Frequent” forms (N=13; each form that
makes up more than 1% of the token file
(N=1,120)
• Frequency has no direct role, but it influences
whether other factors play a role (activation)
and/or how much (amplification)
E&G’s findings
• Activated: morphological regularity, semantic
class, person/number
• Amplified: TMA, switch reference
Theoretical assumption
• Speakers need a certain level of familiarity
with a form to figure out the probabilistic
constraints on variant choice in that context
– It’s an exemplar thing
Bayley (2013)
• Mexican Spanish speakers in Texas, California
• 29 informants, 8676 tokens
• Tested E&G, threw in a couple more factor
groups
• 19 frequent forms (N=2612)
Bayley’s findings
• Amplified: Semantic class (like E&G)
• Weakened: TMA, switch reference,
person/number, lexical aspect (opposite of
E&G)
• So frequency doesn’t activate or amplify
constraints in this data set, contrary to what
Erker & Guy (2012) predict
• Frequency does have a (weak) independent
effect, given a large enough data set
Why replicate again?
• Maybe we can resolve these differences
• We have access to non-contact data
– And contact with English might account for the
wonkiness of E&G and Bayley’s findings
• We have access to Dominican Spanish data
– This variety is famously variable, with higher rates
of overt subjects than elsewhere
– And in more contexts (Martínez Sanz 2011)
The data
• 25 interviews, 2008, Dominican Republic
Data extraction
• 34 speakers
• First 200 tokens per speaker
• 4567 tokens
– after exclusions, other variants, etc.
• 835 frequent-form tokens
Frequent verb forms
Verb form
N
% corpus
% null subject
tengo (I have)
106
2.32
46.23
tenía (I had)
105
2.30
30.48
es (s/he is)
93
2.04
48.39
voy (I go)
93
2.04
36.56
estaba (I was)
80
1.75
28.75
iba (s/he used to go)
67
1.47
47.76
fui (I went)
61
1.33
44.26
sé (I know)
61
1.33
26.23
era (s/he was)
60
1.31
41.67
digo (I say)
55
1.20
47.27
dije (I said)
55
1.20
41.82
Other (linguistic) constraints
• Switch reference
– Bayley & Pease-Alvarez 1996, 1997; Bayley et al. 2012; Cameron
1995, 1996; Flores-Ferrán 2004; Otheguy & Zentella 2012; SilvaCorvalán 1994
• Person/number
– Bayley & Pease-Alvarez 1996, 1997; Cameron 1992, 1996; Erker
& Guy 2012; Flores-Ferrán 2004; Otheguy et al. 2007; Otheguy
& Zentella 2012
• Tense/mood/aspect
– Silva-Corvalán 1994; Bayley & Pease-Alvarez 1997; Bayley et al.
2012
• Ambiguity re person
– Hochberg 1986; Cameron 1993
• Verb semantics
– Travis 2007
Interactions
• Previous studies have coded for different
things
• But many of these overlap
– TMA and form ambiguity
– Semantic features and lexical aspect
– In Bayley, subject type and stativity interact with
frequency
Factor groups coded
• Person/number of the subject
– 1/2/3, singular/plural, plus 2nd indefinite
• TMA of the verb (~ambiguity)
– Preterit, present & related, other
• Semantic features of the verb (~lexical aspect)
• Reference
– Same or switch
• Form frequency
– Tengo and tiene are different forms
Findings
• (See also table in handout!)
• Note that null subject is the application value
Subject Type (Person/Number)
All data
Frequent forms
Infrequent forms
First plural (nosotros/as)
.82
No data
.81
Second + third plural
(ellos/as/ustedes)
Third singular (el/ella/usted/uno)
.82
No data
.81
.51
.66
.48
First singular (yo)
.39
.44
.37
Second singular non-specific
(tu/usted)
Second singular specific (tu/usted)
.19
No data
.19
.12
No data
.11
RANGE
70
20
70
Verb TMA (Ambiguity)
All data
Preterit and
indefinite
Present (and future
and similar)
Imperfect,
conditional,
subjunctive
RANGE
Frequent forms
Infrequent forms
.61
.52
.61
.47
.54
.45
.46
.44
.46
15
10
15
Verb semantics (Lexical Aspect)
All data
Frequent forms
Infrequent forms
Speech act
.55
.63
.54
Motion
.52
.57
.52
Other
.51
.46
.52
Copula
.48
.44
.47
Psychological
.38
.56
.39
Perception
.32
.43
.32
RANGE
23
20
22
Switch Reference
All data
Frequent forms
Infrequent forms
Same referent
.60
.60
.61
Switch referent
.38
.37
.38
RANGE
22
23
23
All tokens together
• All usual factor groups are significant
– Direction virtually identical to earlier (Caribbean
Spanish) studies
• Frequency is not significant
– i.e., higher-frequency items are not more or less
likely to have null subjects
– As in E&G and Bayley (more or less)
Infrequent verb forms
• Behave almost exactly like the full data set
• Because they are the full data set, more or
less!
Frequent verb forms
• Same effects as full data set, very slightly weaker
• One difference: Person/Number:
– Because there were no frequent forms in plurals
or second person, and those were the strongest
constraints in the full data set
– Similar effects in Bayley, Erker & Guy
Frequent verb forms
• TMA (Ambiguity)
-Very slightly weaker effects
• Verb semantics (Lexical aspect)
-very slightly stronger effects
• Switch reference
-No differences
So…
• Frequency alone doesn’t matter
– …although for this variable, nobody has said that
it should
• Frequency doesn’t amplify or activate any
other constraints
Three studies, three different findings
• Why?
1. Speech community differences
• Dialect differences
– Mexican Spanish has less productive overt
subject pronoun system
• Contact
– Previous studies were based on Spanish speakers
in contact with English
– Their constraint systems might be eroded in
infrequent contexts (a Nancy Dorian thing)
2. Data collection differences:
Is this all lexically driven?
• If frequency is determined based on the
corpus, topic and interlocutors could affect
which verb forms are frequent.
• The Bayley interviews are about language
choices, so verba dicendi are frequent
• What’s the theoretical justification for this
method? Surely what matters is which forms
are frequent in a speaker’s cognitive system
3. Similarities across corpora: are
we measuring frequency wrong?
• Even if frequent forms vary from corpus to
corpus, a number of forms appear in all three
corpora.
• In the three data sets, most frequent forms
are in the first and third person singular, and
they include a lot of statives, which trigger
overt subject insertion.
Further research:
redefining frequency
• Person
-Should we look at different Persons separately,
given that they have very different properties?
• Verb Forms and Verb Types
-Frequent verb forms include verbs that bear very
different properties.
-Should we look at how frequency affects verb
types, not just verb forms?
Further research:
redefining frequency
• Frequency might not work the same in phonology
(where most frequency work has been carried out) as
in syntax.
• Given the conflicting results in SPP frequency
studies, we might want to explore different ways of
measuring frequency in syntax (by lemma, form,
collocation?)
Conclusion
• Maybe there’s no frequency effect
• Maybe there is, but we haven’t yet figured
out how to find it
Thanks
• Bob Bayley (and indirectly Greg Guy and
Danny Erker), for sharing work in progress
• The people in the Dominican who shared their
language with Cristina
Contact us
• Gerard: [email protected]
• Cristina: [email protected]