CoChat
Sign Up
Cited by 1,429Language Development and DisordersPhonetics and Phonology ResearchReading and Literacy Development

Statistical learning of tone sequences by human infants and adults

Open access
Saffran, Jenny R, Johnson, Elizabeth K, Aslin, Richard N, Newport, Elissa L·1999·Cognition
DOI 10.1016/s0010-0277(98)00075-4
Open this paper in CoChat — free accountOpen PDF

Abstract

Previous research suggests that language learners can detect and use the statistical properties of syllable sequences to discover words in continuous speech (e.g. Aslin, R.N., Saffran, J.R., Newport, E.L., 1998. Computation of conditional probability statistics by 8-month-old infants. Psychological Science 9, 321-324; Saffran, J.R., Aslin, R.N., Newport, E.L., 1996. Statistical learning by 8-month-old infants. Science 274, 1926-1928; Saffran, J., R., Newport, E.L., Aslin, R.N., (1996). Word segmentation: the role of distributional cues. Journal of Memory and Language 35, 606-621; Saffran, J.R., Newport, E.L., Aslin, R.N., Tunick, R.A., Barrueco, S., 1997. Incidental language learning: Listening (and learning) out of the corner of your ear. Psychological Science 8, 101-195). In the present research, we asked whether this statistical learning ability is uniquely tied to linguistic materials. Subjects were exposed to continuous non-linguistic auditory sequences whose elements were organized into 'tone words'. As in our previous studies, statistical information was the only word boundary cue available to learners. Both adults and 8-month-old infants succeeded at segmenting the tone stream, with performance indistinguishable from that obtained with syllable streams. These results suggest that a learning mechanism previously shown to be involved in word segmentation can also be used to segment sequences of non-linguistic stimuli.

AI snapshot

Written from the paper’s full text.

TL;DR

Adults and 8-month-old infants can use transitional probabilities to segment continuous tone streams, with performance closely matching analogous speech-segmentation tasks. The results suggest a shared statistical learning mechanism for auditory segmentation across linguistic and non-linguistic domains, though the paper limits this to segmentation.

Key findings

  • In Experiment 1, adults segmented tone words from non-words above chance: Language One mean 26.5/36 (74%), t(11)=6.05, P<0.0001; Language Two mean 28.7/36 (80%), t(11)=10.8, P<0.0001; no significant language difference, t(22)=1.26, n.s.; overall tone mean 27.6/36 (77%) did not differ from speech 27.2/36 (76%), t(34)=0.26, n.s.
  • In Experiment 2, adults distinguished tone words from part-words above chance: Language One mean 23.0/36 (64%), t(11)=3.91, P<0.01; Language Two mean 23.7/36 (65.8%), t(11)=5.16, P<0.001; overall tone mean 23.4/36 (65%) did not differ from speech 22.3/36 (62%), t(34)=0.76, n.s.; speech-tone word-score correlation was R=0.939, R-squared=0.88, F(1,4)=29.75, P<0.01.
  • In Experiment 3, 8-month-old infants looked longer at tone part-words (M=6.92, SE=0.48) than tone words (M=5.88, SE=0.45), t(23)=2.14, P<0.05; ANOVA showed a main effect of test item, F(1,46)=9.87, P<0.01, no tone-vs-speech domain effect, F(1,46)=1.9, P=0.17, n.s., and no interaction, F(1,46)=0.13, n.s.
  • Higher transitional-probability words were learned better in Experiment 1: Language One F(5,11)=2.67, P<0.05; Language Two F(5,11)=7.31, P<0.0001; in Experiment 2, part-words containing final syllables of words were falsely chosen more often than initial-syllable part-words, t(23)=2.28, P<0.05.

Methods

  • Design: three experiments using continuous tone streams translated from artificial speech languages; Experiments 1 and 2 used two-alternative forced-choice tests (word vs. non-word; word vs. part-word), and Experiment 3 used infant preferential listening.
  • Sample: Experiment 1 tested 24 non-musician adults; Experiment 2 tested 24 non-musician adults; Experiment 3 tested 24 8-month-old infants (two groups of 12; mean age 7 months 4 weeks; range 7:1 to 8:2), with 18 additional infants excluded for fussiness.
  • Analysis: t-tests, ANOVAs, and correlations compared tone performance with prior speech results; stimuli manipulated transitional probabilities within words versus across word boundaries.

Limitations

  • Stated: the findings are limited to segmentation of sequential auditory materials and may not generalize across modalities or to non-auditory patterns; the results do not definitively eliminate the possibility that tone sequences are processed by a language-specific mechanism.
  • Stated: infant data cannot distinguish transitional-probability learning from interval-sequence learning, and because adult tone languages differed from infant tone languages, interval-based explanations cannot be ruled out for infants.
  • Evident from the text, not framed as author limitations: samples were small, adult participants were restricted to non-musicians, stimuli were artificial continuous streams, and there was no direct neural or mechanistic evidence.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/s0010-0277(98)00075-4

Cites

69 of 69 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

0 of 51 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Overall performance for subjects exposed to Language Two was significantly different from chance: t(11)=5.16, P<0.001.

    “A single sample t-test revealed overall performance significantly different from chance: t(11) = 5.16, P Ͻ 0.001.”
    Quote in the paperSubstantial claim
  • Findingown data

    The mean score for subjects exposed to Language One was 26.5 out of a possible 36 (74%), where chance performance equals 18.

    “The mean score for subjects exposed to Language One was 26.5 out of a possible 36 (74%), where chance performance equals 18.”
    Quote in the paperSubstantial claim
  • Findingown data

    The mean score for subjects exposed to Language Two was 23.7 out of a possible 36 (65.8%).

    “The mean score for subjects exposed to Language Two was 23.7 out of a possible 36 (65.8%).”
    Quote in the paperNot a substantial claim
  • Findingown data

    In the present data, part-words consisting of the final syllables of tone words were incorrectly chosen more frequently than those consisting of the initial syllables of words: t(23)=2.28, P<0.05.

    “An examination of the present data found that the part-words consisting of the final syllables of tone words were incorrectly chosen more fre- quently than those consisting of the initial syllables of words: t(23) = 2.28, P Ͻ 0.05.”
    Quote in the paperSubstantial claim
  • Findingown data

    Four of the six words in Language Two were learned at a level significantly better than chance.

    “Four of the six words were learned at a level significantly better than chance (P Ͻ 0.01 for two words; P Ͻ 0.05 for two words).”
    Quote in the paperNot a substantial claim
  • May combine two claims. 29
  • Classified as probably not a substantial claim. 28
  • The quote only partly states this claim. 1

  • May combine two claims.

    Findingown data

    ANOVAs demonstrated that there were no significant differences in mean scores across words for each language (Language One: F(5,11)=1.19, n.s.; Language Two: F(5,11)=1.84, n.s.).

    “Although ANOVAs demonstrated that there were no significant differences in mean scores across words for each language (Language One: F(5,11) = 1.19, n.s.; Language Two: F(5,11) = 1.84, n.s.)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    There were no significant differences between the two test orders for either Language One (t(10) = 0.694, n.s.) or Language Two (t(10) = 1.87, n.s.).

    “As there were no significant differences between the two test orders for either Language One (t(10) = 0.694, n.s.) or Language Two (t(10) = 1.87, n.s.), data from the two test orders were pooled in the subsequent analyses.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Performance was significantly better on the words of each language containing higher transitional probabilities: Language One: F(5,11) = 2.67, P < 0.05; Language Two: F(5,11) = 7.31, P < 0.0001.

    “As with the linguistic materials used by Saffran et al. (1996a), performance was significantly better on the words of each language containing higher transitional probabilities: Language One: F(5,11) = 2.67, P Ͻ 0.05; Language Two: F(5,11) = 7.31, P Ͻ 0.0001.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Saffran et al., 1996a

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data

    Transitional probabilities between tones within words averaged 0.56 (range 0.33-1.00) for Language Two versus 0.64 (range 0.25-1.00) for Language One.

    “Transitional probabilities between tones within words averaged 0.56 (range 0.33-1.00)versus 0.64 (range 0.25-1.00)for Language One”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The distributions of intervals from the words of Languages One and Two are similar, with ascending minor seconds occurring most frequently in both languages.

    “These distributions are similar, with ascending minor seconds occurring most frequently in both languages.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Transitional probabilities between the tones within words in Language Two averaged 0.71 (range 0.33-1.00), with lower average probabilities across word boundaries (mean = 0.18; range 0.07-0.53).

    “Transitional probabilities between the tones within words averaged 0.71 (range 0.33-1.00),with lower average probabilities across word boundaries (mean = 0.18; range 0.07-0.53).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data

    Transitional probabilities between tones spanning a word boundary averaged 0.15 (range 0.067-0.40) for Language Two versus 0.14 (range 0.05-0.60) for Language One.

    “transitional probabilities between tones that spanned a word boundary averaged 0.15 (range: 0.067-0.40)versus 0.14 (range 0.05-0.60).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Although overall performance was slightly better on Language Two than on Language One, this difference was not significant: t(22)=0.44, n.s.

    “Although overall performance was slightly better on Language Two than on Language One, this difference was not significant: t(22) = 0.44, n.s.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    There were no significant differences between the two randomized test orders for either Language One (t(10)=0.38, n.s.) or Language Two (t(10)=0.80, n.s.).

    “As there were no significant differences between the two randomized test orders for either Language One (t(10) = 0.38, n.s.) or Language Two (t(10) = 0.80, n.s.)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Adult participants in the speech segmentation experiments perceived the synthetic speech stimuli as linguistic, as indicated by their ability to transcribe the speech orthographically and their wonder at how the experimenters were able to remove breaths from the speech.

    “However, the subjective experiences of adult participants in the speech segmentation experiments strongly suggest that these stimuli were perceived as linguistic: when asked, subjects were able to transcribe the speech orthographically, and wondered how the experimenters were able to remove breaths from the speech.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Tone-pairs within words had transitional probabilities of 1.00 and 1.00, whereas tone-pairs within part-words had transitional probabilities of 0.33 and 1.00.

    “Specifically, tone-pairs within words had transitional probabilities of 1.00 and 1.00, whereas tonepairs within part-words had transitional probabilities of 0.33 and 1.00.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    The one tone word that was not learned (ADB) contained a sequence that also occurred spanning a word boundary, but other tone words with this overlap feature were learned significantly above chance.

    “Interestingly, the one tone word that was not learned was the same one (ADB) which contained a sequence that also occurred spanning a word boundary, as discussed in footnote 2. However, other tone words with this overlap feature were learned significantly above chance.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingnone given

    Each word was presented 45 times in the familiarization corpus, whereas the random ordering of words resulted in 15 instances of each part-word.

    “In addition, each word was presented 45 times in the familiarization corpus, whereas the random ordering of words resulted in 15 instances of each part-word.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Only one of the probabilities spanning word boundaries, from A to C# (0.528), overlapped with the within-word probabilities, because this pair of tones also occurred inside the word AC#E.

    “Only one of the probabilities spanning word boundaries, from A to C# (0.528), overlapped with the within-word probabilities, as this pair of tones also occurred inside the word AC#E.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Language learners equipped with the right computational tools should be able to use statistical information to detect word boundaries.

    “To the extent that these statistical cues are available, and they should be available in all languages, given that words are characterized by internal coherence cross-linguistically, language learners equipped with the right computational tools should be able to use statistical information to detect word boundaries.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Humans possess statistical learning abilities which may play an important role in the discovery of word boundaries, and this learning process proceeds automatically as a byproduct of mere exposure.

    “These findings strongly suggest that humans possess statistical learning abilities which may play an important role in the discovery of word boundaries, and that this learning process proceeds automatically as a byproduct of mere exposure.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    All subjects were self-identified as non-musicians, and had not taken instrumental lessons, sung in choruses, or studied music theory since the seventh grade, in order to avoid possible effects of musical expertise on performance.

    “In order to avoid possible effects of musical expertise on performance, all of the subjects were self-identified as non-musicians, and had not taken instrumental lessons, sung in choruses, or studied music theory since the seventh grade.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Learners can detect and use the statistical properties of syllable sequences to discover words in continuous speech.

    “learners can detect and use the statistical properties of syllable sequences to discover words in continuous speech (e.g.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning

    The tone words were identical in their statistical structure to the syllable-based words used in prior experiments, permitting direct comparisons between the acquisition of the same statistical distributions when presented as speech versus tones.

    “Crucially, the tone words were identical in their statistical structure to the syllable-based words used in our prior experiments, permitting direct comparisons between the acquisition of the same statistical distributions when presented as speech versus tones.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisereasoning

    Statistical cues to word boundaries should be available in all languages, given that words are characterized by internal coherence cross-linguistically.

    “To the extent that these statistical cues are available, and they should be available in all languages, given that words are characterized by internal coherence cross-linguistically, language learners equipped with the right computational tools should be able to use statistical information to detect word boundaries.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The statistical structure of the tone words exactly mirrors that of the words used by Saffran et al. (1996a).

    “The statistical structure of these words exactly mirrors that of the words used by Saffran et al. (1996a).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Saffran et al., 1996a

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Subjects were randomly assigned to the two different tone-language groups.

    “Subjects were randomly assigned to the two different tone-language groups.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The individual words were learned similarly across the two domains, suggesting that the particular mode of presentation did not affect which words (or statistical structures) were learned better than others.

    “The individual words were learned similarly across the two domains, suggesting that the particular mode of presentation did not affect which words (or statistical structures) were learned better than others.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A large body of literature focusing on word learning under more naturalistic circumstances also suggests that the ends of words are privileged during acquisition.

    “A large body of literature focusing on word learning under more naturalistic circumstances also suggests that the ends of words are privileged during acquisition (e.g.Slobin, 1973;Echols and Newport, 1992;Golinkoff and Alioto, 1995).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Slobin, 1973

      Not checked in the free run.

    • cites Echols and Newport, 1992

      Not checked in the free run.

    • cites Golinkoff and Alioto, 1995

      Not checked in the free run.

  • Classified as probably not a substantial claim. The quote only partly states this claim.

    Study design and conductnone given

    The transitional probabilities between the new tone and the two tones taken from a word were always zero in the part-words.

    “The transitional probabilities between the new tone and the two tones taken from a word were always zero.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation

    Saffran et al. (1996b) hypothesized that the end-of-word superiority effect might be a side effect of the use of learning mechanisms which compute forward transitional probabilities.

    “Saffran et al. (1996b) hypothesized that the end-of-word superiority effect might be a side effect of the use of learning mechanisms which compute forward transitional probabilities”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Saffran et al. (1996b)

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The 36-item two-alternative forced-choice test was exactly analogous to the test used by Saffran et al. (1996b).

    “To assess learning, we constructed a 36 item two-alternative forced-choice test exactly analogous to the test used by Saffran et al. (1996b).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Saffran et al., 1996b

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Study design and conductreasoning

    If both languages were learned equally well, then performance on the test did not reflect perceptual biases unrelated to the statistical structure of the language.

    “If both languages were learned equally well, then we could be assured that performance on the test did not reflect perceptual biases that were unrelated to the statistical structure of the language.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Two different random orders of the test trials were generated and each was used to test half of the subjects.

    “In both conditions, two different random orders of the test trials were generated and each was used to test half of the subjects.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    The tone words were not constructed in accordance with the rules of standard musical composition, and did not resemble any paradigmatic melodic fragments.

    “The tone words were not constructed in accordance with the rules of standard musical composition, and did not resemble any paradigmatic melodic fragments (e.g., major and minor triads, or familiar three-tone sequences like the NBC television network's chimes).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation

    The final syllable anchors the transitional probability computation that discovers word boundaries, as its frequency serves as the denominator for this computation.

    “the final syllable anchors the transitional probability computation that discovers word boundaries, as its frequency serves as the denominator for this computation.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Saffran et al. (1996b)

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Other aspects of natural language structure cannot be described by finite state regularities and therefore cannot be acquired by a mechanism limited to the computation of transitional probabilities.

    “As Chomsky noted many years ago (Chomsky, 1957), in contrast to word segmentation, other aspects of natural language structure cannot be described by finite state regularities and therefore cannot be acquired by a mechanism limited to the computation of transitional probabilities.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Chomsky, 1957

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Both the parent and the observer listened to masking music over headphones to eliminate bias.

    “Both the parent and the observer listened to masking music over headphones to eliminate bias.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    When subjects engage in pitch processing tasks, right hemisphere activation is present, whereas phonetic processing tasks using the same stimuli produce left hemisphere activation.

    “Moreover, even when phonetic content is available while subjects are engaged in pitch processing tasks, right hemisphere activation, as opposed to the left hemisphere activation resulting from phonetic processing tasks using the same stimuli, is present (Blumstein and Cooper, 1974;Zatorre et al., 1992).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Blumstein and Cooper, 1974

      Not checked in the free run.

    • cites Zatorre et al., 1992 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The present results do not definitively eliminate the possibility that the processing of tone sequences is captured by a language-specific mechanism.

    “However, we recognize that our results do not definitively eliminate the possibility that the processing of tone sequences is captured by a language-specific mechanism.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Infants are presumably computing the transitional probabilities between adjacent tones in the tone stream, grouping tones with high transitional probabilities, and forming tone-boundaries at locations where transitional probabilities are low.

    “That is, infants are presumably computing the transitional probabilities between adjacent tones in the tone stream, grouping tones with high transitional probabilities, and forming tone-boundaries at locations in the tone stream where transitional probabilities are low.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationnone given

    The segmentation of words from fluent speech and tones from tone sequences is a relatively low-level aspect of statistical learning that forms one of the first steps in analyzing these patterned domains.

    “The segmentation of words from fluent speech, and tones from tone sequences, is a relatively low-level aspect of statistical learning which forms one of the first steps in analyzing these patterned domains.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Infants were randomly assigned to hear either Language 1 or Language 2.

    “Infants were randomly assigned to hear either Language 1 or Language 2.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    It is likely that a variety of highly constrained learning mechanisms, at least some of which are specific to humans and to language, will be needed to account for language processing and acquisition as a whole.

    “It seems likely that a variety of highly constrained learning mechanisms, at least some of which are specific to humans and to language, will be needed to account for language processing and acquisition as a whole.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Language acquisition includes many other problems, in addition to segmentation; for example, forming grammatical categories and acquiring hierarchical phrase structure.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    More complex and higher level aspects of language and musical structure will require comparably more complex, and potentially quite different, types of computations to learn them.

    “More complex and higher level aspects of language and musical structure will require comparably more complex (and potentially quite different) types of computations to learn them.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Ongoing work in the authors' labs suggests that the findings are not limited to auditory materials and show many of the same properties for visual pattern learning and visuo-motor sequence learning.

    “However, ongoing work in our labs does suggest that our findings are not limited to auditory materials, and show many of the same properties for visual pattern learning (Asaad, 1998) and visuo-motor sequence learning (Hunt and Aslin, 1998).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Asaad, 1998

      Not checked in the free run.

    • cites Hunt and Aslin, 1998

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Ongoing work suggests that the mechanism for computing transitional probabilities is quite selective in the types of non-adjacent regularities it can compute.

    “but ongoing work suggests that this mechanism is quite selective in the types of non-adjacent regularities it can compute (Calandra, 1998;Newport and Aslin, in press).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Calandra, 1998

      Not checked in the free run.

    • cites Newport and Aslin, in press

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The findings suggesting a common statistical learning mechanism for speech and tone segmentation do not imply either that higher levels of language are acquired by the same mechanisms that perform segmentation or that these mechanisms are domain-general.

    “Our findings suggesting a common statistical learning mechanism for speech and tone segmentation therefore do not imply either that higher levels of language are acquired by the same mechanisms which perform segmentation, or that these mechanisms are domain-general.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The similarity of outcomes between speech and tone streams may pertain only to the segmentation of sequential auditory materials and may not be entirely domain-general.

    “The similarity of outcomes may thus pertain only to the segmentation of sequential auditory materials, and may not be entirely domain-general.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    An additional 18 infants did not complete the experiment due to fussiness.

    “An additional 18 infants did not complete the experiment due to fussiness.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The occurrence of coincidental overlaps between within-word and across-word transitional probabilities makes segmentation more difficult, making subjects' success at it even more notable.

    “The occurrence of such coincidental overlaps makes segmentation more difficult, and therefore subjects' success at it even more notable.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The present results do not directly support or conflict with either the nativist or anti-nativist positions on innate constraints, because they focus only on initial segmentation.

    “As we have noted, the present results do not directly support or conflict with either of these larger positions, since they focus only on one piece of the domains, namely, initial segmentation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The lower performance on part-words compared to non-words is presumably due to the greater difficulty of the part-word test.

    “presumably due to the greater difficulty of the partword test.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The part-word test is more difficult than the word vs. non-word test because words and part-words are more similar.
48 of 48

  • The paper relies on this without support.

    The part-word test is more difficult than the word vs. non-word test because words and part-words are more similar.

    “Subjects were required to discriminate words from part-words, a more difficult task than the word vs. non-word distinction tested in the previous experiment, due to the greater similarity between words and part-words.”

  • The paper relies on this without support.

    Subjects are most likely to false-alarm when the part-word is perceived as highly similar to the material learned during exposure.

    “Subjects are most likely to false-alarm, and incorrectly choose the part-word over the word, when the partword is perceived as highly similar to the material learned during exposure.”

  • The paper relies on this without support.

    The adult tone-languages were sufficiently different from the infant tone-languages that the tone-interval explanation cannot be ruled out for infants.

    “Although our previous analyses suggest that adults do not rely on such tone-interval informa-tion, the adult tone-languages were sufficiently different from the infant tone-languages that this tone-interval explanation cannot be ruled out for infants.”

  • The paper relies on this without support.

    Subjects asked to listen to the speech stimuli and the tone stimuli would not hesitate in labeling the former as language and the latter as music.

    “We have no doubt that subjects asked to listen to the speech stimuli and the tone stimuli would not hesitate in labeling the former as language and the latter as music.”

  • The paper relies on this without support.

    The problem of segmenting elementary units out of a large and apparently unsegmented input is a very general problem, characteristic not only of early levels of language processing, but also of perception in many domains.

    “The problem of segmenting elementary units out of a large and apparently unsegmented input is a very general problem, characteristic not only of early levels of language processing, but also of perception in many domains.”

  • The paper relies on this without support.

    Language acquisition includes many other problems, in addition to segmentation; for example, forming grammatical categories and acquiring hierarchical phrase structure.

    “At the same time, language acquisition includes many other problems, in addition to that of segmentation; for example, forming grammatical categories and acquiring hierarchical phrase structure.”
6 of 6

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Both adults and infants succeeded at segmenting a continuous nonlinguistic tone stream, with performance indistinguishable from that obtained with syllable streams.

    “Both adults and infants succeeded at segmenting the tone stream, with performance indistinguishable from that obtained with syllable streams.”

    From FREQUENCY EFFECTS IN LANGUAGE PROCESSING: A Review with Implications for Theories of Implicit and Explicit Language Acquisition (2002) · cites this paper as Saffran, Johnson, Aslin, and Newport, 1999

  • Research on cognitive development confirms that categorization, statistical learning, and prototype effects are not unique to speech.

    “Research on cognitive development confirms the fact that categorization (83), statistical learning (84), and prototype effects (85) are not unique to speech.”

    From A new view of language acquisition (2000) · cites this paper as [84]

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Consistentt(10) = 0.694reported n.s. · recomputed p = .503

    “t(10) = 0.694, n.s.”

  • Consistentt(10) = 1.87reported n.s. · recomputed p = .091

    “t(10) = 1.87, n.s.”

  • Consistentt(22) = 1.26reported n.s. · recomputed p = .221

    “t(22) = 1.26, n.s.”

  • Consistentt(34) = 0.26reported n.s. · recomputed p = .796

    “t(34) = 0.26, n.s.”

  • Consistentt(10) = 0.38reported n.s. · recomputed p = .712

    “t(10) = 0.38, n.s.”

  • Consistentt(10) = 0.8reported n.s. · recomputed p = .442

    “t(10) = 0.80, n.s.”

  • Consistentt(22) = 0.44reported n.s. · recomputed p = .664

    “t(22) = 0.44, n.s.”

  • Consistentt(34) = 0.76reported n.s. · recomputed p = .452

    “t(34) = 0.76, n.s.”

  • ConsistentF(5, 11) = 1.19reported n.s. · recomputed p = .375

    “F(5,11) = 1.19, n.s.”

  • ConsistentF(5, 11) = 1.84reported n.s. · recomputed p = .185

    “F(5,11) = 1.84, n.s.”

  • Consistentt(22) = 1.02reported n.s. · recomputed p = .319

    “t(22) = 1.02, n.s.”

  • ConsistentF(1, 46) = 1.9reported p = .17 · recomputed p = .175

    “F(1,46) = 1.9, p = .17”

12 of 12

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The stated tone-stream duration does not match its construction

    Major

    Six blocks of 18 three-tone words, with each tone lasting 0.33 seconds and no silent junctures, would last about 107 seconds-not seven minutes. Repeating that stream three times would therefore not provide the stated 21 minutes of exposure.

  • Figure 4 cites a different source for the adult speech results

    Minor

    The results text attributes the adult speech part-word comparison to Saffran et al. (1996b), but the Figure 4 caption attributes those data to Saffran et al. (1996a).

  • Figure 2 cites a different source for the adult speech results

    Minor

    The results text attributes the adult speech comparison to Saffran et al. (1996b), but the Figure 2 caption attributes those data to Saffran et al. (1996a).

3 of 3

Cites

Flagged first, then by citations

Research works printed but not linked: 6

  1. Asaad, P., 1998. Statistical learning of sequential visual patterns. Unpublished senior honors thesis. University of Rochester, New York.
  2. Aslin, R.N., Woodward, J.Z., LaMendola, N.P., Bever, T.G., 1996. Models of word segmentation in maternal speech to infants. In: Morgan, J.L., Demuth, K. (Eds.), Signal to Syntax. Erlbaum, Hillsdale, NJ, pp. 117–134.
  3. Calandra, T.S., 1998. Learning at a distance: statistical learning of non-adjacent regularities. Unpublished senior honors thesis. University of Rochester, New York.
  4. Hayes, J.R., Clark, H.H., 1970. Experiments in the segmentation of an artificial speech analog. In: Hayes, J.R. (Ed.), Cognition and the Development of Language. Wiley, New York, pp. 221–234.
  5. Newport, E.L., Aslin, R.N., in press. The place of statistical learning in a theory of language acquisition. In: Mehler, J., Bonatti, L., Carey, S. (Eds.), Notions of Growth and Development.
  6. Saffran, J.R., 1997. Statistical learning of syntactic structure: mechanisms and constraints. Unpublished doctoral dissertation. University of Rochester, New York.

Cited by

Sorted by citations

Related

Authors
Saffran, J. R., Johnson, E. K., Aslin, R. N., & Newport, E. L.
Affiliations
  • University of Rochester (US)
  • Johns Hopkins University (US)
Corresponding author
Saffran, Jenny R
ORCID
Venue
Cognition
Publisher
Elsevier BV
Published
1999-02
Volume
70
Issue
1
Pages
27-52
ISSN
0010-0277 (ISSN-L)
DOI
10.1016/s0010-0277(98)00075-4
OpenAlex
W2081357278
Semantic Scholar
2d28d075058e4aef8b19c2541028852366e2de3c
Cited by
1,429
References
69
Citations by year
2026: 37 · 2025: 57 · 2024: 40 · 2023: 65 · 2022: 53 · 2021: 61 · 2020: 82 · 2019: 65 · 2018: 65 · 2017: 64 · 2016: 61 · 2015: 58 · 2014: 57 · 2013: 73 · 2012: 101
FWCI
35.84
Type
journal article
Language
en
Open access
Licence
Locations
  • Cognition (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
Funders
  • National Science Foundation (9421064)
  • National Institute for Health and Care Research
  • National Institutes of Health
  • National Institute on Deafness and Other Communication Disorders (DC 00167, R01DC000167)
Topics
  • Social Sciences › Psychology › Developmental and Educational Psychology › Language Development and Disorders
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Phonetics and Phonology Research
  • Social Sciences › Psychology › Developmental and Educational Psychology › Reading and Literacy Development
Keywords
statistical learning, tone words, word segmentation, infants, adults, distributional learning, connected speech, language acquisition
MeSH terms
Adult, Cues, Humans, Infant, Learning, Speech, Statistics as Topic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review