Coda manner of articulation effects Passive Sentences can...
Transcript of Coda manner of articulation effects Passive Sentences can...
1460
1480
1500
1520
1540
1560
1580
1600
1620
Active Passive
RT (ms)
**
Passive Sentences can be Predicted By Adults Karin Stromswold2, Melinh Lai1,2([email protected]), Gwendolyn Rehrig2, Paul de Lacy1
1Linguistics Department, 2Rutgers Center for Cognitive Science
Rutgers, The State University of NJ—New Brunswick
Background Results
Performance distributions
64 targets; at least 40 correct to perform at above chance levels
23/38 participants scored at above chance level; 15 at chance level; none below chance
0
10
20
30
40
50
60
70
80
90
100
Active Passive
% A
ccu
racy
Coda manner of articulation effects
Non-stop Stop
**
NS
Effect of coda voicing & coda articulation
Higher accuracy in active & passive sentences when verb stem had unvoiced coda
Lower accuracy in active sentences when verb stem had stop coda
Effects of voicing and manner of articulation on RT were NS
Questions & Predictions
Temporary syntactic ambiguity between active & passive sentences is resolved by
the inflectional morpheme:
Naïve listeners are able to predict syntax before hearing verb ending in
sentences spoken by a trained phonetician (Stromswold 2002, under review)
Passive verb stems spoken by 7 naïve English speakers were ~52 ms longer
than active verb stems (Rehrig et al, 2015)
Verb stems w/voiced codas ~24 ms longer than verb stems w/unvoiced codas
Verb stems w/non-stop codas ~17 ms longer than verb stems w/stop codas
The pig was kiss -ing the sheep
-ed by the sheep
active
passive
Do naïve listeners use speakers’ verb stem duration to predict active and
passive structures?
If listeners are sensitive to verb durational differences they should be able to
correctly identify active & passive sentences without hearing the verb inflection
How does phonetics affect acoustic cue comprehension?
Coda properties (i.e., voicing and manner of articulation) that affect syllable
duration could either aid or interfere with comprehension of verb stem duration
Materials & Methods
Stimuli
Audio recordings of a participant from Rehrig et al.’s production study
32 actives: The _____ was _____-ing the _____.
32 passives: The _____ was _____-ed by the _____.
Nouns
bear cat cow dog
duck elephant fox frog
hippo kangaroo lion monkey
mouse pig rabbit rhino
sheep turtle zebra
Verbs (grouped by coda properties)
Voiced Unvoiced
kick lick
Sto
p
comb pat pinch poke
scrub punch tickle
touch trap
chase kiss
Non
-
sto
p
shove push wash
Discussion
References
Listeners may use verb stem duration as a cue to predict passives
Participants identified passive sentences faster & more accurately, regardless
of phonetic properties of verb stem coda
Verb stem lengthening as a probabilistic cue
Active sentences more prevalent than passive sentences
Participants may update probabilistic distributions of potential structures after
listening to verb stem duration
Would require stored knowledge of “normal” stem duration for all verbs
Comparison of stem duration in utterance to “normal” duration stored in
lexicon informs distribution of probabilities of active/passive structures
Unclear whether acoustic cues exist for just passives or for actives also
Presence of acoustic cue for actives would have resulted in higher accuracy
But correctly identified active sentences clearly had shorter stems
Interactions of phonetics & syntax more significant in active sentences
Stem length does at least inform their decision during the task
Perhaps participants could not fully decide after hearing actives
Just noticeable difference in passive stimuli
Durational differences of ~20% are perceptually important (Klatt 1976)
Active cue in stimuli may not have been shortened enough to be noticeable
Phonetic effects of coda
Phonetics of coda may interfere with comprehension of acoustic cue
Lower accuracy in voiced passives & non-stop actives
Listeners may be unsure whether stem duration due to syntax or coda effects
Participants may be listening specifically for vowel duration (as opposed to the
entirety of the verb stem)
Vowels shorter if followed by unvoiced consonants (Umeda 1975; Klatt 1976)
Actives may get additional cue from unvoiced coda due to shortened vowel
No prior established effects of manner of articulation on vowel duration
Unclear what drives higher accuracy in actives with non-stop codas
Investigation of robustness across phonetic effects is ongoing
Acknowledgments
1. Stromswold, K., Kharkwal, G., Sorkin, J. E., & Zola, S. (Manuscript submitted for publication). Tracking the elusive passive: The processing of spoken passives., 28-40.
2. Stromswold, K., Eisenband, J., Norland, E., Ratzan, J. (2002). Tracking the Acquisition & Processing of English Passives: Using Acoustic Cues to Disambiguate Actives & Passives. Conference talk presented at CUNY 2002.
3. Rehrig, G., Beier, E., Chalmers, E., Schrum, N., Stromswold, K. (2015). Robust acoustic cues indicate upcoming structure in active and passive sentences. Poster presented March 19th 2015, The 28th Annual CUNY Conference on Human Sentence Processing, Midtown Los Angeles Radisson Hotel, Los Angelos, CA.
4. Klatt, D. H. (1976). Linguistic uses of segmental duration in english: Acoustic and perceptual evidence. Journal of the Acoustical Society of America, 59(5), 1208-1221.
5. Umeda, N. (1975). Vowel Duration in American English. Journal of the Acoustical Society of America, 59(2), 434-445.
This work was generously supported by the Aresty Foundation and the National Science Foundation (REU CNS-1062735; IGERT DGE 0549115). Special thanks to Nora Beier, Liz Chalmers, and Nic Schrum for conducting the production experiment that this study followed upon, and Aldo Mayro and Rachel Shu for confirming their findings. Additional thanks to my colleagues in the Language Processing and Development Lab at Johns Hopkins University and to Akira Omaki for supporting my attendance. The greatest thanks go to my advisers, Karin Stromswold and Paul de Lacy for their guidance and insight throughout the entirety of this study.
Overall results: Active vs. Passive
Participants were more accurate and faster at predicting passive sentences
Stem length & prediction accuracy
Verb stems were 29 ms shorter on correct active trials and 25 ms longer on correct
passive trials
Interaction: F(1, 37) = 12.471, p = 0.001 Interaction: F(1, 37) = 7.44, p = 0.01
Interaction: F(1, 61) = 31.75, p < 0.0001
0
10
20
30
40
50
60
70
80
90
100
Active Passive
Accuracy (%)
***
F(1,37) = 54.387
p < 0.001
F(1,37) = 8.765
p = 0.005
0
10
20
30
40
50
60
70
80
90
100
Active Passive
% A
ccu
racy
Coda voicing effects
Voiced Unvoiced
***
*
F(1,37) = 44.136
p < 0.001
F(1,37) = 9.785
p = 0.003
250
275
300
325
350
Active Passive
Mea
n d
ura
tio
n (
ms)
Stem duration differences in participants'
correct & incorrect trials
Correct Incorrect
***
***
F(1,61) = 22.095
p < 0.001
F(1,61) = 18.17
p < 0.001
F(1,37) = 6.451
p = 0.015 Comprehension gating experiment
1) Fixation (500 ms)
2) Audio recording truncated at offset of verb stem coda plays
3) Forced choice between
fully written active & passive
sentences
-Key presses corresponding
to L and R sentences
e.g., “The pig was kiss—”