Samenvattingen lezingen over eerste- en tweedetaalverwerving, 18 juni 2004, MPI.
Georganiseerd door de Nederlandse Vereniging voor Fonetische Wetenschappen
Voice and Language Discrimination by Dutch-learning Infants
Elizabeth Johnson* & Ellen Westrek**
*Max Planck Institute for Psycholingusitics, Nijmegen
**Free University Amsterdam
Linguists have argued that all languages belong to one of three rhythmic categories: stress-timed (e.g. English and Dutch), syllable-timed (e.g. Italian and Spanish), and mora-timed (e.g. Japanese). Acoustic studies have provided evidence for the physical reality of these three rhythmic categories, and psychological studies have demonstrated that language rhythm plays an important role in language acquisition and word segmentation.
Language discrimination studies with English-learning infants have revealed that infants younger than 2 months can hear the difference between two languages only if the two languages belong to different rhythmic classes. In other words, cross-category language pairs (e.g. Italian and Japanese) are discriminated whereas within-category language pairs (e.g. Dutch and English) are not. By 5 months, within-category discrimination is possible only if one of the within-category languages is the native language (e.g. English and Dutch). Thus, both rhythmic information as well as language familiarity play a role in early language discrimination.
In this talk, I will discuss language discrimination by Dutch-learning infants. Using the Switch paradigm, infants were habituated to three voices speaking Language 1, and tested on a fourth voice speaking Language 1 and a fifth voice speaking Language 2. In Experiment 1 (Dutch versus Japanese), infants dishabituated to Language 2 in the test phase regardless of which language they were habituated to. However, they only dishabituated to the new voice in Language 1 if Language 1 was Dutch. In Experiment 2 (Japanese versus Italian), infants once again dishabituated to Language 2 regardless of which language they were habituated to. Infants in both conditions failed to dishabituate to the new voice in Language 1. In Experiment 3, infants are being tested on two stress-timed languages: German and Dutch. Current results support the hypothesis that language rhythm and familiarity play an important role in voice and language discrimination by Dutch-learning infants.
Word segmentation from continuous speech: an ERP study with 10-month-old infants.
Valesca Kooijman.*, Peter Hagoort* & Anne Cutler**
*FC Donders Centre for Cognitive Neuroimaging, University of Nijmegen,
**Max Planck Institute for Psycholinguistics, Nijmegen.
In their first year of life, before they start to speak, infants rapidly learn to deal with the sound structure of their native language. An important step in this first year of language acquisition is learning to recognize words in continuous speech (word segmentation).Behavioral studies have shown that infants learn to segment words from continuous speech between approximately 7 and 10 months of age. The predominant stress pattern of a language seems to play an important role in acquiring this skill. To study this step in language acquisition in more detail, we designed a new auditory ERP repetition paradigm. In this paradigm we presented 10-month-old infants with 20 blocks of words and sentences, while measuring EEG. Each block consisted of a Familiarization phase and a Test phase. In the Familiarization phase we presented the infants with 10 tokens of the same two-syllable word with stress on the first syllable. In the Test phase, which followed immediately after familiarization, we presented infants with eight randomized sentences of which four contained the familiarized word. The remaining four sentences contained a new two-syllable word, also with stress on the first syllable. The Test phase shows an effect of Familiarity in the form of a negative-going deflection on the familiar words as compared to the unfamiliar words in the sentences. This effect starts well before the end of the critical word. This result shows that 10-month-old infants can indeed segment words from continuous speech. Moreover, the timing of the effect shows us that they need approximately only the first half of the word to do so.
Tania S. Zamuner
University of Nijmegen
Dutch allows both voiced and voiceless obstruents in initial position, but only voiceless obstruents in final position. The underlying /d/ surfaces as [t] in the plural noden ‘emergencies’, but as [t] in the singular nood ‘emergency’. Voiceless obstruents do not undergo a similar alternation: [t] is produced in both nooten ‘nuts’ and noot ‘nut’. To acquire knowledge of voicing neutralization, the learner needs to realize that in final position voiced stops are phonotactically illegal.
Previous research has shown that by 9-months, English-learning infants have acquired knowledge about their language specific phonotactics (e.g. Juscyzk et al., 1995). We predicted that by 9-months, Dutch-learning infants should demonstrate knowledge of voicing phonotactics. In the first studies, 9- and 11-month-old Dutch-learning infants were presented with lists of non-words ending in phonotactically legal voiceless obstruents versus phonotactically illegal voiced obstruents. Infants showed no preference for either list.
We then explored the possiblity that infants were not sensitive to the nature of the phonotactic pattern tested. While voiced stops are restricted in final position, they do occur in the language; therefore, this requires a sophisticated knowledge of phonotactics. Similarily, the voicing contrast may not be as salient as other contrasts. 9-month-old Dutch-learning infants were then presented with a more salient contrast: lists of non-words ending in native phonemes versus non-native phonemes. Infants showed no preference for either list. We then tested whether infants showed no preference for voicing or native phonotactics because contrasts occured in final position. 9-month-old Dutch-learning infants were presented with lists of non-words beginning with native phonemes versus non-native phonemes. Infants listened significantly longer to non-words begining with native phonemes phonemes. The combined results suggest that infants are not very sensitive to phonotactic patterns in final position. Moreover, the findings suggest that infants learn phonotactics in different prosodic positions at different times in development.
Early lexical representations: evidence from perception and production.
University of Nijmegen
Central in our investigation are the shape of early lexical representations. What do early words look like? Are they stored with detailed phonetic information, or are they more abstract and perhaps holistic at first? Although most researchers favor the idea that words are stored with phonetic detail (Werker & Fennell 2004, Swingley 2004), the evidence so far is inconclusive.
Our research on early speech production has led us to believe that representations are not phonetically detailed at first. Rather, they start out being holistic, and become segmentalized in the course of development. Once words are segmentalized, children may generalize over the phonological characteristics of their lexical forms, which is the beginning of an abstract phonological system consisting of wellformedness or markedness constraints. Moreover, there are certain significant asymmetries with respect to place of articulation: coronal is behaving differently from other places of articulation, and hence assumed to be underspecified.
These claims have repercussions for early word perception. If words are stored as highly abstract units, then the prediction is that for word recognition detailed phonetic information is not used either. If certain features are underspecified, asymmetry in behavior is expected to show up in perception as well. To test these claims, we replicated and expanded previous research by Werker and colleagues (1997, 2001, 2004, to appear). In addition, we collected and analyzed production data and CDI scores for both production and perception. The results of our (still ongoing) investigations are highly interesting.
In a first experiment using the switch paradigm, we tested whether children prefer to listen to the switch condition when tested with newly learned words (bin or din). Like the English infants, Dutch 14 month old infants do not listen significantly longer to the new words than to the ‘old’ words, indicating that the difference between bin and din is not picked up. In a second experiment we used the switch procedure to test the difference between bon and don. In this case, infants listened significantly longer to the new forms than to the ‘old’ ones, suggesting that the bon-don contrast is perceived. It is, however, not as simple as that. In both experiments, children listened significantly longer to the new words if they are able to produce an initial labial-coronal contrast, which suggests that the child’s productive lexicon indeed plays a role.
Analogical effects on past-tense formation in L1 and L2 learners of Dutch
Pim Mak* & Mirjam Ernestus**,*
*University of Nijmegen
**Max Planck Institute for Psycholinguistics
There is a simple deterministic rule for past-tense formation in Dutch. If the verbal stem in the infinitive ends in an unvoiced obstruent, the simple past suffix is -te. In all other cases, it is -de. Ernestus & Baayen (2003,2004), however, showed that adult speakers of Dutch do not simply follow this rule. They tend to construct past tense forms by analogy with phonologically similar words in the lexicon. When analogy supports the suffix prescribed by the rule (congruent verbs), adults make less mistakes in the choice of the suffix then when analogy and the rule are in conflict (incongruent verbs). In this talk, we will present a study on the production of past-tense forms by primary school children, both L1 and L2 learners. If the difference between the congruent and incongruent verbs (the congruency effect) indeed depends on patterns of analogy in the lexicon, it should become greater when a child’s vocabulary becomes larger. Dutch primary school children at the end of grades 5 to 8 (after 3 to 6 years of reading) were asked to construct the past-tense forms of congruent and incongruent verbs. The experiment showed that the congruency effect is already present in grade 5. Also, the effect becomes larger over the years. Children whose native language is not Dutch show a smaller congruency effect, which forms additional support that the congruency effect depends on vocabulary size.
The notion of language proficiency from an aural processing perspective
Amsterdam Center for Language and Communication (ACLC), University of Amsterdam
Theories of first and second language acquisition must seek to explain the following two phenomena. First, whereas L1 acquisition appears to lead to full native proficiency for all learners (i.e., all mentally healthy children with normal hearing), L2 acquisition only very seldom appears to lead to native or nativelike proficiency. Second, despite the fact that L1 learners all appear to reach full native proficiency, big differences between them exist. According to Hyltenstam and Abrahamsson (2003), both biological factors (the so-called critical period) and social-psychological factors play a role in early as well as late language acquisition; however, with increasing age of onset, the impact of maturational factors diminishes, while the impact of social-psychological factors increases. Using the theory of Hyltenstam & Abrahamsson as a starting point, I will propose that we must distinguish between a core and a non-core language proficiency. The core pertains to the recognition of high-frequency words in normal speech, is acquired via a process of implicit learning, and is attainable for all L1 and L2 learners. The non-core involves forms of explicit learning, and is therefore affected by a range of social-psychological factors (intelligence, working memory capacity, vocabulary size, education). In this talk, I will present the three hypotheses, elaborate on the underlying theoretical notions, and suggest a method of how to test them empirically.
Dutch listeners' processing of word-final voicing in English
Max Planck Institute for Psycholinguistics
Whereas Dutch has voiced and voiceless obstruents, only voiceless obstruents occur in word-final position. In a series of experiments we investigated Dutch listeners' processing of the voice distinction in word-final position, and its use for the recognition of words in English. In a categorization experiment Dutch and English participants categorized /b/-/p/, /d/-/t/, /z/-/s/, and /v/-/f/, in nonword-initial and -final position. Dutch listeners were found to categorize voiced and voiceless obstruents in word-final position as accurately as in word-initial position, and as accurately as English listeners. To investigate to which extent Dutch listeners use this ability for the recognition of words, English words with an obstruent in word-final position were recorded in two ways: unaltered, and with final voiced consonants replaced with voiceless ones and vice versa. Thus 'globe' became 'glope', and 'cheap' became 'cheab'. The replacement of final consonants did not result in existing words, but in so-called 'near-words'. Real words and near-words were presented to Dutch and English participants in a categorization experiment. Dutch listeners were found to rely more on lexical cues than English listeners: Dutch listeners gave more 'b' responses to the final consonant in 'glope' and more 'p' responses to 'cheab' than English listeners. When the same items were presented in an auditory lexical decision task, Dutch listeners misjudged the near-words as real words more often than English listeners, again relying on the lexicon. To investigate how strongly near-words activate the real word, the same items were used as auditory primes in a bimodal priming experiment, with the real words as visual targets. For English listeners, near-words did not prime real words. For Dutch listeners, 'glope' primed 'globe' as strongly as 'globe' did. 'Cheab', however, did not prime 'cheap'. Thus Dutch listeners were tolerant of incorrect devoicing, but intolerant of incorrect voicing of word-final obstruents.
The development of L2 categorization in multiple scenarios
Phonemic and phonetic mismatches between the sound categories of different languages frequently lead to problems in L2 acquisition. Following the simple, and widely accepted, assumption of L1 transfer, it is predicted that we will find at least three different learning scenarios for the categorization of L2 sounds, namely starting with fewer sound categories, starting with more categories and starting with similar but not equal categories. I will present different L2 categorization cases that illustrate each of the three scenarios: The learning of the English vowels /i/ and /I/ by Spanish speakers will be shown to represent the first scenario, the second scenario will be illustrated by the learning of the Spanish vowels /i/ and /e/ by Dutch listeners, and the learning of the Canadian French vowels /ae/ and /E/ by Canadian English speakers will be shown as an example of the third scenario. For each scenario, two arguments about the staring point will be put forward. First, it is argued that the L2 starting point equals the cross-language categorization of the vowel sounds at hand (e.g. the native Spanish categorization of English /i/ and /I/ or the Dutch native categorization of Spanish /i/ and /e/). And second, it is argued that the learner is faced with a different L2 learning problem in each of the scenarios, namely learning to categorize more vowels, learning to categorize less vowels and learning to re-categorize vowels. As for solving the L2 learning problems, following the, perhaps, controversial assumption of availability of L2 development, it will be argued that the L2 learners can modify their initial L2 categorization in order to gradually approximate native categorization. Crucially, it is claimed that the learners of each scenario will adjust their L2 vowel categorization through different developmental (sequence of) strategies. The results of cross-language and L2 categorization experiments will be shown to confirm the hypothesised L2 scenarios as well as the hypothesised L2 developments.
From phoneme to lexicon in non-native listening
Anne Cutler*, Andrea Weber**, Takshi Otake***
*Max Planck Institute for Psycholinguistics
When phonemic contrasts of first and second languages mismatch, native phonemic categories can capture second-language input. This effect can have far-reaching consequences for understanding:
1. Spurious word activation. Broersma (2002) showed that Dutch listeners accepted non-words such as chass and lem as English words (chess, lamb); such sequences can occur in spoken English (chastise, lemon), possibly activating more competitor words for non-native than for native listeners. (See also Broersma's abstract.)
2. Pseudo-homophony. Homophones like [mi:t] must be disambiguated via context (let's meet, grill meat). If failure to distinguish contrasts induces homophony, then non-native listeners will have to do such disambiguation via context more often than native listeners.
3. Temporary ambiguity. Weber and Cutler (2004) tested Dutch listeners to English in an eye-tracking experiment, and found that they fixated longer on distractor pictures with names containing vowels confusable with vowels in a target name (pencil, given target panda) than on less confusable distractors (beetle, given target bottle). English native
listeners did not do this.
However, Weber and Cutler also found that the confusability was asymmetric: given pencil as target, panda did not distract more than distinct competitors. They suggested that stored representations may maintain second-language distinctions even when native phonemic categories effectively over-rule the distinctions in input processing.
A subsequent experiment tests for such asymmetry with Japanese listeners' perception of English r/l contrasts. We also tested for asymmetry in non-natives' pseudo-homophony, via a lexical decision study examining repetition priming. English materials, presented to
Dutch and Japanese listeners, included pairs such as cattle/kettle and right/light. Dutch listeners responded significantly faster to one member of a cattle/kettle pair after having heard the other member earlier (compared with having heard a control word), suggesting that both words had been activated whichever had been heard. Japanese listeners, however, showed no such priming for cattle/kettle words, but did show repetition priming across r/l pairs (e.g. right/light).