This page gives an overview of my research, which is in the area of speech perception development. In all studies, I attempt to combine at least two approaches. My favourite approaches are corpus analyses of language input, experimental speech perception research and computer modeling.
COPYRIGHT NOTICE: The documents distributed here have been provided as a means to esure timely dissemination of scholarly and technical work on a noncommeral basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder
Please go to the published source for citations and quotes. Do not cite unpublished work without explicit permission.
Paul Boersma.
It is assumed that infants learn about the phonemes from their native language by performing distributional learning on the acoustic values in their input. Although infants can do this in laboratory tasks, it is unknown whether the speech input infants receive has the correct properties for distributional learning.In this project we therefore set out to investigate whether the contrast between Dutch /A/ and /a:/ can be learned from infant-direceted speech, and whether either spectral quality, duration, or a combination of both cues is necessary to recover the category structure.
Distributional learning was modelled using a mixture-of-gaussians, as well as with a symmetric neural network with competitive learning. Both methods were used, as the former is used most frequently in the literature to answer such questions, whereas the latter is a more realistic implementation of what may happen in infants' brains.
Preliminary results show that two cues are necessary to recover /A/ and /a:/ from input. Learning based on only one cue fails in this respect. The results thus support the notion that distributional learning could underly phoneme acquisition, but add that correlated cues may be crucial.
This research was conducted within my Toptalent-project and Paul Boersma's Vici-project, at the University of Amsterdam.
+Titia Benders & Paul Boersma. (2012, Jun).
Learning from multiple acoustic cues for phoneme acquisition: Dutch infants' perception, language input, and neural networks.
International Child Phonology Conference, 4-6 June 2012, Minneapolis, USA. abstract.
+Titia Benders & Paul Boersma. (2012, Jun).
Learning from multiple acoustic cues for phoneme acquisition: Language input, neural networks, and Dutch infants' perception.
MARCS Monday Meeting, 14 May 2012, University of Western Sydney, Sydney, Australia.
+Titia Benders & Paul Boersma. (2012, Mar).
Learning from multiple acoustic cues for phoneme acquisition: Dutch infants' perception, language input, and neural networks.
NETwerk Eerste Taalverwerving, March 2012, Amsterdam. abstract.
Dorothy Mandell.
Dutch /A/ and /a:/ differ in two acoustic cues, spectral quality and duration. Dutch native adult listeners weigh the spectral quality heavier when they have to categorize ambiguous instances in between these tokens. Infants, on the other hand, are thought to be very sensitive to the salient vowel duration cue. In the present research we investigated whether Dutch infants are sensitive to the native cue (quality) or the salient cue (duration) when categorizing ambiguous sounds as /A:/ or /a/.
To this end, we implemented a categorization method suitable for infants. In the task, participants first learn the association between the sounds /A/ and /a:/ and an interesting outcome at the opposite ends of the screen. Subsequently, the ambiguous sounds /A:/ and /a/ are presented as stimuli and it is investigated where the infants predict the outcome to appear.
Preliminary results show that adults and 15-month-old infants, but not 9-month-old infants can do the task. The behaviour of the first two group is highly similar and suggests they differentially treat the ambiguous sounds, /A:/ and /a/. For the infant group, the pupil dialation disambiguates their behavior.
This research was conducted within my Toptalent-project, at the University of Amsterdam.
+Titia Benders & Dorothy Mandell. (2011, Apr).
The integration of acoustic dimension in infants' phoneme perception: Introducing and testing a categorization paradigm.
Society for Recearch in Child Development (SRCD), 31 March-2 April, Montréal, Canada. abstract.
Paul Boersma.
Dutch /A/ and /a:/ differ in two acoustic cues, spectral quality and duration. The objective of this study was to find out whether infants use one of these cues, both cues, or a combination of both cues when discriminating between these vowels. To that end, we tested infants in a sequential preference procedure in which they would hear series of 8 sounds with alternations between /A/ and /a:/, alternations between vowels differing only in duration (e.g., /A/-/A:/), alternations between vowels differing only in spectral quality (e.g., /A/-/a/), and series without any alternation (e.g., /A/-/A/). Infants' looking time was recorded by the eye-tracker.
Results show that infants discriminate between /A/ and /a:/ when both cues are signal the contrast. The contrast signalled by two cues is more salient to the infants tha the contrast signalled by a single cue, which undermines the idea that infants may be using a single cue to discriminate between these two vowels. We interpret these results as evidence that Dutch infants have learned that /A/ and /a:/ typically differ in two acoustic cues.
This research was conducted within my Toptalent-project, at the University of Amsterdam.
+Titia Benders. (2012, Jan).
Dutch infants’ discrimination of /A/ and /a:/: evidence for simultaneous learning of multiple cues.
Old World Conference in Phonology, 18-21 January 2012, Berlin, Germany. abstract.
+Titia Benders (2011, Jun).
Dutch infants' disrimination of /A/ and /a:/: development into the second year of life.
International Child Phonology Conference, 15-18 June 2011, York, UK. abstract.
This project was made possible by the indispensable help of my research assistants: Gisela Govaart, Marieke van den Heuvel, and Maartje ter Hoeve.
Across languages, parents speak differently to their babies than to other adults. The exact way in which parents adapt to their infants is, however, language specific, and also influenced by the infants' age and sex as well as the parents' own sex. To investigate Dutch IDS, I set out to collect a corpus of Dutch speech of mothers and fathers who interacted with their infant of 9, 11, or 15 months of age. Parents were instructed to play with a set of toys containing the corner vowels /i/, /u/, /a:/, and /A/, and the fricative /s/. This method allowed for analyzing both the suprasegmental and segmental aspects of Dutch IDS.
The first results were obtained from a longitudinal subset of the corpus, with recordings of 18 mothers playing with their infant at 11 and 15 months. The results show that Dutch infant-directed intonation is similar to what is found in IDS in other languages, although the pitch movements appear rather small. More importantly, Dutch mothers make their vowels sound more affective when speaking to their infant. As a side-effect of this affective speaking style, the vowel space in Dutch IDS is smaller than in adult-directed speech, whereas in many languages the vowel space is larger in IDS. The infants' age also influences the mothers' productions, as the vowels are produced more affectively to the 11-month-olds, and the pitch adaptations are larger in speech to the 15-month-olds.
Coding of a subset of recordings from the corpus with both mothers and fathers and an equal number of boys and girls is underway, to investigate the effect of the infants' and parents' sex on Dutch IDS.
This research was conducted within my Toptalent-project at the University of Amsterdam.
+Titia Benders. (2011, Dec).
Een akoestische beschrijving van de hoekklinkers in babygerichte spraak in het Nederlands.
Dag van de Fonetiek, 15 December 2011, Utrecht, Netherlands. abstract.
MA project, supervised by Suzanne Curtin, Daniel Swingley and Paola Escudero.
Previous work by Suzanne Curtin, Chris Fennell and Paola Escudero suggested that 15-month-old Canadian-English infants are not sensitive to changes in the vowels' second formant (F2) in a word learning task. However, this task was conducted in the Switch paradigm, in which infants might miss slight mispronunciations in phonetic properties of the signal that they do represent. The aim of the present study was therefore to investigate whether infants may be sensitive to F2 in a different word recognition paradigm.
In our study, we teach children one word for one novel object. They are also familiarized with a second novel object, but do not learn a word for this object. At test, infants see the two objects side-by-side and are requested to look at the [learned word]. We expect them to look at the novel object they had learned the word for. The crucial test is where infants will look if they hear a slight mispronunciation of the learned word in F2. If they look less at the novel object they had learned the word for, that suggests that they notice that the word is mispronounced and thus that they are able to process F2 in novel word learning.
Results are still preliminary. However, it appears that children are not sensitive to F2 whereas they do notice changes in F1. The results thus confirm those of Curtin, Fennell and Escudero in a different paradigm.
This research was conducted at Suzanne Curtin's Speech Development Lab at the University of Calgary.
+Titia Benders, Suzanne Curtin, Paola Escudero, Daniel Swingley. (2010, Mar).
Fifteen-Month-old Infants' Sensitivity to Vowels' First and Second Formant in Novel Word Learning
International Conference on Infant Studies, 2010, Baltimore, US. Within the workshop: Signals and Representations: What Different Sound Contrasts Tell us About Phonological Development. abstract.
+Titia Benders. (2009, Jan).
Fifteen-month-old infants’ sensitivity to vowels’ first and second formant in novel word learning.
Anela Juniorendag, University of Amsterdam.
+Titia Benders. (2008, Dec).
Tonghoogte en articulatieplaats in de woordrepresentaties van 15 maanden oude baby’s.
Nieuwsbrief Werkverband Amsterdamse Psycholinguisten, jaargang 26, nr. 4.
+Titia Benders. (2008, Oct).
Fifteen-month-old infants’ sensitivity to vowels’ first and second formant in novel word learning.
Baby Circle, Leiden University.
+Titia Benders & Paola Escudero. (2008, Jan)
First and second formant in infants' novel words.
Infant Language Lab, Department of Psychology, University of Pennsylvania.
+Titia Benders. (2007, Nov).
Phonetic detail in infants' novel words.
IFA-talk, University of Amsterdam.
Paola Escudero and Matthias Sjerps.
When listening to speech (or when experiencing any sensory stimuli, really), listeners are sensitive to what they've heard before. In a speech sound categorization task, this is for example observed as the stimulus range effect: The effect that the boundary between two categories is on a higher value along the continuum if all stimuli have a relatively high value along the continuum, whereas the boundary between the same two categories is on a lower value if all stimli are shifted towards the lower end of the continuum. In the present study we investigated whether the low-level stimulus range effect is influenced by the number of response options a listener considers in the task. The results of our study show that listeners are more subject to the stimulus range effect if they can choose from two response options, than if they can choose from 5 response options, even if those extra three responses do not occur during the experiment.
+Titia Benders, Matthias Sjerps, & Paola Escudero. (Published on-line). On the multitude on influences during a single vowel categorization response.
Journal of the Acoustical Society of America, 131:4.
Copyright (2012) Acoustical Society of America. This article may be downloaded for personal use only. Any other use requires prior perission of the author and the Acoustical Society of America. The article appeared in Journal of the Acoustical Society of America, 131:4, and may be found at link.
+Titia Benders, Paola Escudero & Matthias Sjerps. (2010, Nov).
The interrelation between the stimulus range and the number of response categories in vowel categorization.
2nd Pan-American/Iberian Meeting on Acoustics, 15-19 November 2010, Cancun, Mexico. abstract.
+Benders, T. & Escudero, P. (2010).
The interrelation between the stimulus range and the number of response categories in vowel categorization.
Proceedings of Interspeech 2010, Makuhari, 27-30 September 2010.
+Titia Benders & Paola Escudero (2010, Sep).
The interrelation between the stimulus range and the number of response categories in vowel categorization.
11th Annual Conference of the International Speech Communication Association, (Interspeech), 27-30 September 2010, Makuhari, Japan. proceedings article.
Paola Escudero and Karin Wanrooij.
We have investigated whether distributional learning may help Spanish L2-learners of Dutch to improve their categorization of a difficult vowel contrast. Distributional learning was tested in a condition with the peaks of the distribution on the average values of the Dutch vowels, as well as on more extreme values. The latter condition was envisioned to mirror the enhanced way in which American-English mothers speak to infants (but see above that Dutch mothers don't do this). Results show that after distributional learning, the Spanish learners are indeed better at categorizing a difficult L2 vowel contrast, especially if they learned from the enhanced distribution.
This research was conducted within Paola Escudero's Veni-project, at the University of Amsterdam.
+Paola Escudero, Titia Benders, & Karin Wanrooij. (2011). Enhanced bimodal distributions facilitate the learning of second language vowels.
Journal of the Acoustical Society of America, 130:4, EL206-EL212.
Copyright (2011) Acoustical Society of America. This article may be downloaded for personal use only. Any other use requires prior perission of the author and the Acoustical Society of America. The article appeared in Journal of the Acoustical Society of America, 130:4, and may be found at link.
Paola Escudero (PI) and Silvia Lipski (PI).
Being a native speaker of a language causes one to listen to speech sounds in a language-specific manner. Learning a new language may then be difficult, because certain acoustic differences are not perceived as relevant anymore. In our behavioral study, we found that Dutch native listeners rely mostly on spectral quality to categorize the vowels /A/ and /a:/, whereas Spanish L2-learners of Dutch rely mostly on duration to tell these two vowels apart. Similar differences between the L1 and L2 listeners were found in a pre-attentive EEG task, which did not require overt responses.
This research was conducted within Paola Escudero's Veni-project, at the University of Amsterdam, as well as at the University of Konstanz.
+Silvia Lipski, Paola Escudero & Titia Benders. (Accepted).
Language experience modulates weighting of acoustic cues for vowel perception: an event-related potential study.
Psychophysiology.
+Paola Escudero, Titia Benders, & Silvia Lipski. (2009).
Differences in the perceptual cue weighting of Dutch vowels by Dutch, German and Spanish natives.
Journal of phonetics, 37:4, 452-465.
Paul Boersma.
Prototypes (best examplars, not the representational equivalent some scholars adhere to) of speech sound categories occupy a location in a highly multidimensional space. Finding the prototype by having listeners rate all possible exemplars in the multidimensional space becomes exponentially more time-consuming with every dimension that is added. We therefore set out to develop an algorithms to guide listeners through the multi-dimensional space towards the best exemplar. The algorithm performs local hill-climbing with decreasing step sizes. The partipant in the experiment provides the input to the algorithm in deciding on every step in which direction in the multidimensional space the algorithm should move. Computer simulations showed that our algorithm was more robust to decision noise on the part of the listeners than an earlier algorithm developed by Iverson and Evans.
This research was conducted within my Toptalent-project, at the University of Amsterdam.
+Titia Benders & Paul Boersma. 2009.
Comparing methods to find a best exemplar in a multidimensional space.
Proceedings of Interspeech 2009, Brighton, 6-10 September 2009.