Cambridge Books Online

Cambridge Books Online
http://ebooks.cambridge.org/
The Cambridge Handbook of Child Language
Edited by Edith L. Bavin, Letitia R. Naigles
Book DOI: http://dx.doi.org/10.1017/CBO9781316095829
Online ISBN: 9781316095829
Hardback ISBN: 9781107087323
Paperback ISBN: 9781107455504
Chapter
8 - Crosslinguistic perspectives on segmentation and categorization in
early language acquisition pp. 159-182
Chapter DOI: http://dx.doi.org/10.1017/CBO9781316095829.008
Cambridge University Press

8
Crosslinguistic
perspectives on
segmentation and
categorization in early
language acquisition
Barbara Höhle
8.1 Introduction
The previous chapter has shown how infants’ speech perception is shaped
by the developing phonological system and how this process leads to the
establishment of lexical representations and the processing of content
words. The present chapter will pursue the issue of the interactions
between innate processing capacities and the specific requirements of
the language to be learned with a special focus on crosslinguistic research
including the initial steps infants take to enter the particular morphosyn-
tactic system of the target language. Though there is an increasing body of
research on bilingual children in this area, this overview will be restricted
to studies with monolingual infants.
One of the fascinating questions of language acquisition research con-
cerns the nature of the interplay between the innate abilities the child
brings to solving the task of learning a language and the impact of the
different conditions of experience provided by the child’s exposure to one
or more language(s) and their specific structural features. Language acqui-
sition is a developmental process in which the target is subject to multiple
I thank Jürgen Weissenborn for his longstanding cooperation in our common research on the early acquisition of
function words and for his comments on an earlier version of this contribution. My research cited in this chapter was
supported by several grants from the German Science Foundation (DFG HO 1960/5-1/2; HO 1960/6-2; HO1960/8-1).
Last but not least I thank my colleagues from the Special Research Cluster Information Structure (SFB 632). The
opportunity for cooperative work within this framework sharpened my view on crosslinguistic variation and the necessity
of incorporating it into acquisition research.
Downloaded from Cambridge Books Online by IP 157.181.88.58 on Wed Mar 16 13:00:04 GMT 2016.
http://dx.doi.org/10.1017/CBO9781316095829.008
Cambridge Books Online © Cambridge University Press, 2016
160 BARBARA HÖHLE
variations. Even though the crosslinguistic variation can be described

within a restricted set of dimensions or parameters that constrain the
grammatical options a language can have (Chomsky 1981b), we have to
assume that the learning mechanisms involved are characterized by at
least some flexibility to cover this variation. The enterprise of crosslinguis-
tic language acquisition research initiated by Slobin and his colleagues
(Slobin 1985b and subsequent volumes) has demonstrated that, to a cer-
tain degree, different kinds of languages present different kinds of acquisi-
tion tasks to the child. As a consequence we see that specific structural
features of the language to be learned have an impact on the acquisition
process from very early on. Nevertheless, it is far from clear how and when
the child – equipped with some kinds of universal mechanisms to acquire
a language – adapts to the specific problems that a language poses in the
acquisition process.
The flexibility of the learning mechanism and the variation in the type
of information that these mechanisms rely on will be the focus of this
chapter. Looking at two tasks that the child has to master and seems to
master within the early phases of language acquisition – namely the
segmentation of the speech input into linguistically relevant units and
the assignment of these units to syntactic categories – we will see that
learners use various cues to solve these problems. An overview of existing
data on language processing and language learning capacities in children
within the first two years of life will show that there is no unique trajectory
of language acquisition across languages but that the path is shaped by
specific features of the target language from early on.
8.2 Some methodological remarks
Because we are examining an early phase of language acquisition compris-

ing mainly the first eighteen months of life, we will mainly present experi-
mental data from studies on young children’s speech perception using one
of the established methods used in this type of research (for an overview,
see Hoff 2012). Most of the studies that will be discussed have used the
headturn preference paradigm. Some others – especially those that have
studied newborns – were run with the high-amplitude sucking paradigm.
Nevertheless, even when the same experimental paradigm has been used,
there is still a great deal of variation in the methodological details of the
studies, which makes it very hard to compare the results across different
studies. The outcome of experiments with infants may be heavily influ-
enced by small experimental variations, including the number of trials
used, the duration of the familiarization phase if included, the number of
different stimuli. Studies using the headturn preference paradigm have
found familiarity effects (i.e. longer listening times to familiar stimuli) as
well as novelty effects (i.e. longer listening times to unfamiliar stimuli) in
Crosslinguistic perspectives on segmentation and categorization 161
experiments with very similar setups (Thiessen & Saffran 2003). This might
be the result of an interaction involving the complexity of the stimuli
presented and the child’s developmentally changing capacity to process
them, a capacity that has not yet been understood in its full complexity
(see Burnham & Dodd 1999, Houston-Price & Nakai 2004). Nevertheless,
according to the model by Hunter and Ames (1988), a phase of a familiarity
preference might be followed by a phase of novelty preference within one
single child across the trials of an experiment. This raises the problem in
determining novelty preferences, familiarity preferences or null effects
when listening times are averaged over all trials of an experimental ses-
sion. More recently, experiments using event-related potentials (ERPs; see
Chapter 4) have entered the research on language development in early
infancy. In the areas considered in this chapter, some studies on word
segmentation have used this technique, which does not depend on a
behavioural response by the infants and thus may be especially applicable
to studying young infants. However, due to brain and processing matura-
tion, ERP responses in infants do not necessarily mirror those found in
adults and understanding these differences is still a challenge.
Comparing the performance of children across languages is an enter-
prise that requires a high degree of methodological comparability of the
experiments with respect to the kind of stimuli, the age of the children and
the details of the experimental procedure. Our review will show that, so
far, only a few studies are genuinely crosslinguistic in this sense. However,
as most of the studies on early language acquisition have been conducted
with English-learning children, special attention will be paid to research
on languages that are different from English.
8.3 Crosslinguistic issues in word segmentation
8.3.1 Rhythmical typology and rhythmical sensitivity

So far, the typological approach has been followed most consistently in
research on the emergence of word segmentation capacities in children
learning stress-timed and syllable-timed languages. Traditionally, stress-
timed languages (e.g. most of the Germanic languages) are considered to
base their rhythm on the recurrence pattern of stressed syllables while
syllable-timed languages (e.g. most of the Romance languages) base their
rhythm on the syllable itself (Pike 1945). More recent approaches search-
ing for acoustic correlates of this classification in the speech signal suggest
that the perception of rhythmic differences across languages may depend
on the temporal characteristics of the vocalic and consonantal portions of
the signal (Ling, Grabe & Nolan 2000, Ramus, Nespor & Mehler 1999).
Stress-timed languages tend to have smaller and more variable propor-
tions of vocalic parts and a larger variability in the duration of intervocalic
consonantal sections than syllable-timed languages. These properties
162 BARBARA HÖHLE
reflect the differences between languages with respect to their typical

syllable structure. Stress-timed languages have complex and variable syl-
lable structures ranging from simple CV syllables to syllables with com-
plex consonant clusters in the onset and coda. In contrast, syllable-timed
languages show less variable syllable patterns with a dominance of simple
CV syllables, leading to a high proportion of vowels and high homogeneity
of the syllable structures observed. Adult listeners are able to perceive
differences between languages based on exactly these cues (Ramus,
Dupoux & Mehler 2003).
But what about children? Infants’ ability to discriminate between differ-
ent languages seems to reflect exactly the boundaries set by these rhyth-
mical groupings. Nazzi and colleagues (Nazzi, Bertoncini & Mehler 1998)
systematically tested the ability of newborns to discriminate languages of
the respective types. They found that French newborns discriminate
between languages of different rhythmic groups (e.g. English from
Japanese or English from Italian) but not between languages of the same
rhythmic group (e.g. English from Dutch or Italian from Spanish). Hence,
infants seem to be equipped with perceptual mechanisms that are sensi-
tive to the phonetic features that constitute the rhythmical structure of
language. Using different types of synthesized Dutch and Japanese speech
strings, Ramus (2002) demonstrated that these discrimination capacities
are, in fact, dependent on the rhythmic properties of speech and not on
general intonation patterns.
This sensitivity to rhythmical information seems to be the basis for a
rapid acquisition of at least some of the specific rhythmic or prosodic
features of the target language. At the age of 5 months, English-learning
infants already show a high sensitivity to the rhythmical features of their
native language. Even though they are still not able to discriminate foreign
languages belonging to the same rhythmical class, they can discriminate
their native language from other languages from the same rhythmical
class (Nazzi, Jusczyk & Johnson 2000). The observation of very early acqui-
sition of prosodic features of the target language is supported by data
showing that German infants as young as six months prefer to listen to
trochaic (i.e. strong-weak) as compared to iambic (i.e. weak-strong) disylla-
bic words, while French infants of the same age do not show the same
behaviour even though they can discriminate the rhythmical patterns
(Höhle, Bijeljac-Babic, Herold, Weissenborn & Nazzi 2009). Language-
specific asymmetrical brain responses to trochaic and iambic items,
which differ between German and French 4-month-olds, probably reflect
an even earlier impact of language experience on the processing of the
rhythmical patterns (Friederici, Friedrich & Christophe 2007). In addition,
the ability to discriminate stress patterns seems to be weakened in French
infants towards the end of the first year of life: 9- to 10-month-old French
babies only discriminate stress patterns after longer stimulus exposure
than required by 6-month-olds (Bijeljac-Babic, Serres, Höhle & Nazzi 2012),
or in segmentally less variable strings than their Spanish-learning age

mates (Skoruppa et al. 2009). This attenuation persists to adulthood as
French adults’ stress and rhythm processing varies more strongly depend-
ing on acoustic and segmental properties of the speech signal than that
of Spanish or German adults (Bhatara, Boll-Avetisyan, Unger, Nazzi &
Höhle 2013, Dupoux, Pallier, Sebastián-Gallés & Mehler 1997, Dupoux,
Peperkamp & Sebastián-Gallés 2001) – a finding that may be related to
the fact that French, unlike German or Spanish, does not have contrastive
lexical stress.
Infants acquiring Spanish – a syllable-timed language with a more equal
distribution of trochaic and iambic disyllabic words than German – were
not found to demonstrate a trochaic listening bias at 9 months when the
words only contained CV syllables (Pons & Bosch 2010). However, the
infants were sensitive to the correlation of stress and syllable weight in
Spanish, preferring disyllabic items in which a heavy CVC syllable was
stressed compared to a condition in which syllable weight and stress did
not coincide. For English, a language that is rhythmically and prosodically
very similar to German, a preference for listening to the trochaic rhythmic
pattern was only found for 9- and not for 6-month-olds (Jusczyk, Cutler &
Redanz 1993). The age discrepancy between German- and English-learning
infants may reflect either a delay for English learners due to some crucial
differences between the two languages, or methodological differences
between the studies. While the study with German- and French-learning
infants used simple CVCV sequences that showed only prosodic and no
segmental variation, the study with English learners used a set of different
English trochaic and iambic words. However, taken together the results
reported in this section suggest that infants are equipped with fine-tuned
mechanisms to perceive prosodic properties of speech which are attuned
quickly to the rhythmical and stress properties of the ambient language.
The following section will discuss how this development feeds into seg-
mentation of the speech stream into linguistically relevant units.
8.3.2 Using rhythm to segment speech

Adapting the metrical segmentation strategy initially proposed for
speech processing in adults (Cutler, Mehler, Norris & Segui 1986) to
language acquisition, many researchers have proposed that the rhythmic
sensitivity of infants plays a crucial role in determining a segmentation
strategy for the detection of word boundaries in the native language
(Curtin, Mintz & Christiansen 2005, Echols, Crowhurst & Childers 1997,
Houston, Juscyzk, Kuijpers, Coolen & Cutler, 2000, Jusczyk, Houston &
Newsome 1999, Morgan & Saffran 1995, Nazzi & Ramus 2003, Nazzi,
Iakimova, Bertoncini, Frédonie & Alcantara 2006). In stress-timed lan-
guages, there is a coincidence of the boundaries of metrical feet and of
words. The initial boundary of a metrical foot – defined by a strong
164 BARBARA HÖHLE
syllable – is a reliable cue for an initial word boundary for a reasonable

number of content words in these languages (Cutler & Carter 1987).
For studying word segmentation in infancy, a classic experimental setup
using the headturn preference paradigm has been widely used (Jusczyk &
Aslin 1995). In this procedure infants are exposed either to isolated words
or to passages that include repetitions of a target word during a familiar-
ization phase at the beginning of the experiment. When the child has been
familiarized with isolated words, passages in which these words occur and
passages in which they do not are presented during testing. When the
child has been familiarized with a passage, isolated words that occurred in
the passage or words that did not occur in the passage are presented during
testing. Longer listening to familiarized words or to a passage containing
the familiarized words is considered as evidence for a successful segmen-
tation of the target words. More recently, specific ERP components related
to the recognition of familiar words have also been reported (e.g. Goyet,
de Schonen & Nazzi 2010, Kooijman, Hagoort & Cutler 2005, Männel &
Friederici 2013, and see Chapter 4).
Infants learning English, German or Dutch – which all are stress-timed
languages with a predominance of trochaic feet – have been found to use a
segmentation strategy that is adapted to this correlation of stress and word
boundaries (Höhle 2002, Houston et al. 2000, Jusczyk, Houston & Newsome
1999, Kooijman, Hagoort & Cutler 2009, Männel & Friederici 2013, Polka &
Sundara 2012). Learners of these languages between 7 and 10 months
succeeded in segmenting disyllabic words with initial strong stress out of
continuous speech. However, there seems to be a developmental delay
for detecting words with an initial weak syllable (Jusczyk, Houston &
Newsome 1999). This suggests that at the beginning of word segmentation
a metrical strategy applies that assumes strong syllables to be word-initial,
and attaches subsequent adjacent weak syllables to the strong one. This is
exactly what the hypothesis of a rhythmically triggered segmentation
strategy would predict for children learning a language in which the
trochaic foot is the basic rhythmical unit.
For a full evaluation of the metrical segmentation hypothesis, data from
languages other than stress-timed ones are necessary. So far, French has
been the primary language of comparison. Nazzi and colleagues (Nazzi
et al. 2006) have provided evidence for a delay in disyllabic word segmenta-
tion in French infants. Only at the age of 16 months – but not at 8 or
12 months – were French learners able to detect disyllabic words in
passages when they had been familiarized with these words in isolation.
These findings are in line with the assumption that French infants start
word segmentation based on the syllable, which allows a correct segmen-
tation of monosyllabic words, but then further segmentation cues are
needed to be successful with disyllabic words. However more recent data
are at odds with this developmental pathway. Using the same method as
in the Nazzi et al. (2006) study, Polka and Sundara (2012) found that
8-month-old infants learning Canadian French can successfully segment

disyllabic words independently of whether the stimuli are presented in
their native Canadian dialect or in European French. A follow-up with
European-French-learning infants in Paris (Nazzi, Mersad, Sundara,
Iakimova & Polka 2014) indeed yielded evidence that the emergence of
segmentation skills is not equivalent in infants learning these two dialects:
although Parisian 8-month-olds were found to be able to segment disylla-
bic words in this study, they could only do this when familiarized with
passages and then tested on isolated words, and their performance was
compromised when presented with Canadian French stimuli. As argued by
Nazzi et al. (2014), the procedure using the passage familiarization and
isolated word testing may be less challenging than the word familiariza-
tion and passage testing procedure, and may allow infants to profit more
from statistical cues during the repeated presentation of the passages
during the familiarization. With this reasoning, and given the fact that
Canadian-French-learning infants were successful in the word familiariza-
tion order, the word segmentation skills of Parisian 8-month-olds seem to
be less robust than those of Canadian 8-month-olds, the reasons for which
may be found in prosodic differences between these two dialectal variants
of French. These crossdialectal differences show that even slight variations
across languages may have an impact on early word segmentation skills,
thereby challenging the hypothesis that the assignment of a language to a
rhythmical group is the most relevant factor that predicts which kind of
segmentation routine infants initially use.
Further crosslinguistic differences between stress- and syllable-timed
languages become obvious when infants’ reactions to single syllables are
considered. In their original study, Jusczyk, Houston and Newsome (1999)
tested English infants’ detection of the strong syllables in disyllabic words
either by familiarizing them with stimuli containing a single syllable
corresponding to the strong syllable of the disyllabic words (e.g. tar from
guitar or king from kingdom) and then testing them with material contain-
ing the complete disyllabic words, or vice versa. Seven-month-olds did not
show evidence of detecting this syllable when it occurred in a trochaic
word, but they did when it occurred in an iambic word, while 10-month-
olds showed no reactions to the isolated syllables at all. These findings
corroborate the notion that the younger infants treat the trochaic, but not
the iambic, words as units while the older ones treat both word types as
coherent units. In contrast, Parisian 12-month-olds and Canadian 8-month-
olds have shown some evidence of responsiveness to the isolated initial
and final syllables in addition to the whole disyllabic word (Nazzi et al.
2006, Polka & Sundara 2012). This suggests that a stronger sensitivity to the
single syllables of a disyllabic word is retained in French-learning as com-
pared to English-learning infants.
Additional evidence for a more important role for syllables in the speech
processing of infants learning a syllable-timed language comes from
166 BARBARA HÖHLE
studies that tested the segmentation of monosyllabic words. According to

results by Bosch, Figueras, Teixidó and Ramon-Casas (2013), Catalan- and
Spanish-learning 6-month-olds can segment monosyllabic words from
passages. In contrast, English-learning infants have shown this capacity
by the age of 7 months but not yet at the age of 6 months (Jusczyk & Aslin
1995).
So far, the reported research suggests that at least in stress-timed lan-
guages, rhythmic properties may provide important cues for early word
segmentation and shape the relevant skills from early infancy. However, if
rhythmic structure guides their initial segmentation, how do English,
German or Dutch infants find out that their language has a predominance
of trochaic feet, i.e. that they should segment their input in strong–weak
rather than weak–strong patterns? Although the majority of languages
with a foot organization have trochaic feet as their dominant rhythmical
pattern, there are a considerable number of languages in which iambic
feet are dominant (Goedemans & van der Hulst 2013). This observation
leads to a ‘chicken and egg’ problem: to find out whether the stressed
syllable indicates the beginning or the end of a unit in a given language, is
it not the case that the infant must first find some of the relevant units in
the input before he or she can establish what the dominant pattern is (see
Thiessen & Saffran 2003)?
One solution to this problem may come from indications that different
acoustic parameters are associated with strong syllables at the beginning
or at the end of a unit – a regularity which is called the iambic-trochaic law
(Hayes 1995). According to this law, initial strong syllables are marked by a
higher intensity (i.e. are louder) than the surrounding syllables while final
strong syllables are marked by a higher duration (i.e. are longer). There is
consistent evidence that adults’ perception of strings of syllables or non-
speech sounds that alternate in either intensity or duration corresponds to
this law: sequences alternating in intensity are typically grouped in
strong–weak pairs while an alternation of duration leads to a weak–strong
grouping, with some indications that the force of the law can be modu-
lated by language experience (Bhatara, Boll-Avetisyan, Unger, Nazzi &
Höhle 2013, Hay & Diehl 2007, Iversen, Patel & Ohgushi 2008).
If this perceptual bias is innate, infants must be equipped with a general
auditory mechanism that may guide their perception of strong syllables as
either marking the beginning or the end of units and thus their grouping
of the input into either trochaic or iambic feet. Existing data are still
scarce, but thus far infants appear to be sensitive to differences between
the relevant acoustic cues. While intensity and similarly pitch information
led English 6-month-olds and Italian 7-month-olds to a trochaic grouping
(Bion, Benavides-Varela & Nespor 2011, Hay & Saffran 2011), the same age
groups did not show any use of durational information, which was, how-
ever, evident in English 9-month-olds. Yoshida et al. (2010) found that
slightly younger English-learning 7- to 8-month-olds treated duration as a
cue for iambic grouping, but Japanese 7- to 8-month-olds did not respond
to the durational cues – which is consistent with findings for Japanese
adults (Iversen et al. 2008). Taken together, the available data suggest that
pitch and intensity may already be used in accordance with the iambic–
trochaic law by rather young infants while the use of durational cues
seems to be subject to a longer developmental process and thus is a
stronger candidate for being a result of language exposure. However, the
earlier exploitation of intensity and pitch cues fits with the initial advan-
tage of segmenting trochaic words in English-learning infants as it should
be possible to identify their first strong syllable by these cues. So far,
however, the applicability of the iambic–trochaic law to the segmentation
of words from natural speech is unclear: all experiments reported used
sequences with manipulated acoustic cues in such a way that only dura-
tional, pitch or intensity variation was present. In natural language, a
coalition of acoustic cues which often includes duration is typically asso-
ciated with stressed syllables independently of their position (e.g. Sluitjer
& van Heuven 1996 for Dutch, Jessen, Marasek, Schneider & Clahßen 1995
for German, Klatt 1976 for English). Hence, it is possible that the predic-
tions made by the iambic–trochaic law are less suitable for word segment-
ing than for the segmentation of larger prosodic units (Nespor et al., 2008).
8.3.3 Further phonological cues to word boundaries

Assuming that the availability of rhythm or other prosodic information
to signal word boundaries varies across languages, the question arises as
to which other word-inherent phonological cues may support segmenta-
tion (see Chapter 7). In general, infants are highly proficient in comput-
ing information concerning the position and the segmental context in
which the sound segments of their language typically occur within words
or syllables (Jusczyk, Luce & Charles-Luce 1994). This allows them to
make use of allophonic variation between segments at word boundaries
(Johnson & Jusczyk 2001, Jusczyk, Hohne & Bauman 1999) as an indica-
tion of word boundaries. Furthermore, phonotactic information may
play a role: Dutch 9-month-old infants can distinguish between words
with phonotactically illegal consonant clusters and words with legal
consonant clusters depending on their position in the word and use
this information for segmentation (Friederici & Wessels 1993). Dutch and
English 9-month-olds recognize the language-specific positional restric-
tions on segments and the language-specific phonotactic constraints of
their respective languages (Jusczyk, Friederici, Wessels, Svenkerud &
Jusczyk 1993). In addition, phonotactic probabilistics (i.e. the frequency
of co-occurrence of segments within words and across words) have been
shown to be an efficient cue for segmentation by 9-month-old English-
learning infants (Mattys & Jusczyk 2001a, Mattys, Jusczyk, Luce & Morgan
1999).
168 BARBARA HÖHLE
Again, one could argue that making use of these properties calls for prior
computation of the phonotactic constraints or the phonotactic probabil-
ities of the ambient language and would require the previous segmenta-
tion of the relevant units like syllables or words. However, the sensitivity
of infants to these co-occurrence patterns may be based on their highly
efficient mechanisms for tracking transitional probabilities between adja-
cent segments, which by themselves seem to play a crucial role in segmen-
tation processes (Saffran, Aslin & Newport 1996). These statistical learning
mechanisms may have the capacity to segment speech based only on the
computation of distributional properties of segments and syllables.
In addition to transitional probabilities across adjacent segments, more
recent research has provided evidence that infants even in their first year
of life can also track non-adjacent dependencies in their input (Friederici,
Mueller & Oberecker 2011, Gervain & Werker 2012). Non-adjacent depen-
dencies across consonants and vowels may also provide useful informa-
tion for segmentation processes. One instance is the typical order of
consonantal segments in words. The so-called labial coronal law describes
a bias found across many languages according to which the sequencing of
a labial followed by a coronal (e.g. bata) is more frequent in words than the
reversed order (e.g. taba). More recently it has been shown that the labial
coronal law also affects speech perception in young children. French
10-month-olds prefer to listen to words that follow the labial coronal
law compared to words with the reversed order of these consonants
(Gonzalez-Gomez & Nazzi 2012, Nazzi, Bertoncini & Bijeljac-Babic 2009).
Gonzalez-Gomez and Nazzi (2013) found that these non-adjacent phono-
tactic patterns also affect word segmentation: French 10-month-olds were
only successful in segmenting CVC words when the order of consonants
followed the labial coronal law and not when it was violated. At the age of
13 months the segmentation of words violating the law was still disadvan-
taged compared to that of words conforming to the law.
Less is known about the role of vowels in infants’ word segmentation.
Results from Italian adults suggest that non-adjacent dependencies across
vowels may be less important for word segmentation than the same type of
dependency across consonants (Toro, Nespor, Mehler & Bonatti 2008).
However, this effect may be language-specific as there is evidence that
adult speakers of languages that have vowel harmony make use of this cue
to segment their speech input (Suomi, McQueen & Cutler 1997). In a series
of experiments van Kampen, Parmaksiz, van de Vijver and Höhle (2008)
tested Turkish infants’ sensitivity to vowel harmony. In Turkish, all vowels
within one word have to belong to one and the same of two different
harmony classes based on the front–back distinction, with front vowels
forming one class and back vowels forming the other class (see Kabak &
Vogel 2001). If two syllables with vowels not belonging to the same har-
mony class appear in adjacent syllables, there is a very high probability
that there is a word boundary between the syllables. Van Kampen et al.
(2008) first tested Turkish-learning 6-month-olds’ sensitivity to this feature

of their language by presenting them with disyllabic items that either only
contained harmonic vowels or that violated the harmony. The Turkish
infants preferred to listen to the stimuli that obeyed the Turkish harmony
restrictions. In contrast, German 6-month-olds – learning a language with-
out vowel harmony – did not show any listening preferences for the same
harmonic or non-harmonic sequences. This suggests that the preference
observed for the Turkish infants is not due to general acoustic preferences
for sequences of vowels with similar articulatory features but is the result
of their exposure to a language that systematically uses vowel harmony in
its lexical inventory.
A further experiment tested Turkish 9-month-olds’ exploitation of this cue
for word segmentation. The infants were familiarized with a disyllabic har-
monic word either in a non-harmonic context in which the word was pre-
ceded by a non-harmonic syllable or in a harmonic context in which it was
preceded by a harmonic syllable. In the test phase only the disyllabic words
were presented. Longer listening times for the words that had been familiar-
ized in the non-harmonic context were obtained, suggesting that the infants’
segmentation was supported by the harmony disruption in the sequences
that contained a non-harmonic preceding syllable. This finding adds to the
existing literature showing that infants acquire knowledge about the typical
word forms in their language rather fast and use different types of cues
provided by their ambient language to find a solution to the segmentation
problem.
In general, the findings reported in this and the preceding section show
that infants may use a number of cues that are inherent in lexical words to
find a solution for initial word segmentation, cues that may be based on
rhythmic and prosodic properties or on segmental properties related to
the distribution of consonants, vowels or syllables. The next section will
show that distributional information beyond the word boundary also plays
a crucial role and that especially the class of functional morphemes may be
important for segmenting speech from early on.
8.4 Functional morphemes and their role in segmenting

the speech stream
In many languages, functional morphemes are essential for indicating the

grammatical structure of utterances. They occur as function words – also
called free functional morphemes – like determiners or auxiliaries, or as
bound functional morphemes like inflectional endings for case or number.
Although the emergence of functional morphemes is delayed compared to
that of lexical morphemes in children’s spontaneous language production,
research during the last decades has shown that children process these
elements from early infancy on. Given their typical high frequency of
170 BARBARA HÖHLE
occurrence (Shi, Morgan & Allopenna 1998) and infants’ known sensitivity
to frequently occurring strings in the signal, it can be assumed that the
first representations of these elements are established rather early during
language development and that these elements can then serve in a top-
down fashion as ‘anchor points’ (Valian & Coulson 1988) in infants’ sub-
sequent language processing and acquisition.
8.4.1 Free functional morphemes

One of the first experiments that provided evidence that very young infants
can detect function words within continuous speech was conducted by
Höhle and Weissenborn (2003). Using the classic experimental design to
study word segmentation in infants (Jusczyk & Aslin 1995, see above),
6-month-old and 7- to 8-month-old German infants were familiarized with
different function words and other unstressed closed class elements includ-
ing determiners as well as prepositions. After the familiarization they were
presented with passages either including one of the familiarized items or
not. The 7- to 8-month-olds but not the 6-month-olds showed longer listen-
ing times to passages including a familiarized function word than to pas-
sages not including a familiarized item. According to these results, the older
infants had detected the crucial elements in continuous speech despite the
fact that they had the typical features of unstressed functional morphemes
in continuous speech; for instance, showing only half of the duration of the
corresponding words presented in isolation. This suggests that – at least for
German learners – there might be fewer perceptual disadvantages for
unstressed functional items than previously thought.
These findings found crosslinguistic support from data from even
younger Canadian-French-learning infants. Shi, Marquis and Gauthier
(2006) did a similar experiment in which 6-month-olds were familiarized
with one of two determiners (either la ‘definite article, singular feminine’
or des ‘indefinite article, plural’) and then tested with noun phrases includ-
ing the familiarized determiner or the non-familiarized one. Infants
showed a familiarity effect for the phrases including the familiarized
determiner. A further experiment in which two phonetically highly simi-
lar functors (la ‘the’ and ta ‘your’) were used during the test phase, failed to
show an enhanced attention to the phrase containing the familiarized
determiner. This suggests that the representations that the 6-month-olds
build up for the word forms during the familiarization phase are not fully
phonetically specified. It is not clear whether the fact that French learners
reacted to the function words at a younger age than the German children
tested by Höhle and Weissenborn (2003) is due to methodological differ-
ences or whether it reflects systematic differences in the speed of the
acquisition processes in the two languages. A crucial difference between
the two studies is the complexity of the stimuli presented during the test
phase. While Shi and colleagues tested with isolated noun phrases that
always included the critical determiner in initial position, Höhle and

Weissenborn used whole sentences; thus, the crucial elements were
embedded in longer strings with material preceding and following them.
This may have rendered their detection harder than in the material used
by Shi and colleagues.
These experiments show that French- and German-learning infants have
the perceptual ability to detect free functional morphemes. The recognition
of at least some high-frequency functional morphemes seems to be present
not much later. Shi, Werker and Cutler (2006) showed that 11- to 13-month-
old English-learning infants, but not 8-month-olds, preferred to listen to
sequences consisting of combinations of a real determiner or pronoun and a
nonsense word (e.g. the breek, his tink) compared to sequences in which the
functor had been replaced by a nonsense syllable (e.g. ris tink). Comparable
findings are reported by Hallé, Durand and de Boysson-Bardies (2008) for
French-learning 11-month-olds.
Further results suggest that babies are not only able to recognize these
elements in continuous speech but that their occurrence supports infants’
segmentation of adjacent words. Höhle and Weissenborn (2000) familiar-
ized German-learning 11-month-olds with disyllabic sequences, either
representing a noun phrase including the definite article plus a monosyl-
labic noun (e.g. der Kahn ‘the boat’) or a disyllabic word, the first syllable of
which did not constitute a word form by itself and the second syllable of
which was segmentally identical to the noun of the noun phrases (e.g.
Vulkan ‘volcano’). In the test phase, passages were presented in which only
the strong syllable of the familiarized items appeared in new contexts. The
children who had been familiarized with the noun phrases responded with
longer listening times to the passages including this syllable, but not the
children familiarized with the disyllabic words. This suggests that the
children had segmented the noun phrases during the familiarization but
had represented the disyllabic words as one unit. Since the only difference
between the familiarization strings was in the form of the first syllable, we
assume that the children – based on an already existing form representa-
tion of the determiner – had segmented this item from the string and were
left with a second monosyllabic item. Nine-month-olds did not show this
effect.
A comparable effect for even younger English learners was obtained
by Shi, Cutler, Werker and Cruickshank (2006). They familiarized 8- and
11-month-olds with nonsense words (e.g. breek) preceded by a higher (the)
or a lower (her) frequency function word or by nonsense syllables that were
phonetically very similar to the real function words (kuh, ler). In the test
phase of the experiment only the nonsense words were presented to test
their recognition. The 11-month-olds showed longer listening times to
a nonsense word that had been familiarized together with the higher-
frequency existing determiner (the) as compared with the nonsense
functor. For the lower-frequency function word no effect was observed.
172 BARBARA HÖHLE
The result pattern of the 8-month-olds was different. They showed longer
listening times to those nonsense words that had either been familiarized
with the or with kuh than to those familiarized with the lower-frequency
functor or its phonetic foil. This suggests that both age groups recognize
the higher-frequency functor as a familiar string in the input and therefore
seem to segment the string before the nonsense word, which facilitates the
recognition of the item in the test phase. While the phonological repre-
sentation of the higher frequency function word the seems to be already
quite specific for the 11-month-olds, it is still underspecified for the
8-month-olds, leading to the same results for the real and the nonce
function word.
Similar results were reported by Shi and Lepage (2008) for infants learn-
ing Canadian French. They familiarized 8-month-olds with sequences of
either the French indefinite plural determiner des or the first-person pos-
sessive pronoun mes or a nonsense syllable kes together with an infrequent
French noun. In the test phase infants were only presented with the
isolated nouns. The infants listened longer to those nouns that had been
familiarized with one of the existing function words than to the nouns that
had been presented with a preceding nonsense syllable during familiariza-
tion. To test for frequency effects of the functors used, Shi and Lepage ran a
second experiment in which the pronoun mes was replaced by the less
frequent second-person possessive pronoun vos. This form did not yield the
same effect that had been observed for the more frequent pronoun mes in
the first experiment.
The results from these three different languages converge in showing
that infants establish representations of free functional morphemes
throughout their first year of life and that these representations affect
their processing of speech and enhance the recognition of a following
element as an independent unit. The providing of boundary information
for the adjacent word and thereby the supporting of the segmentation of
the speech stream could be one of the sources of this facilitation effect.
Furthermore, the results suggest that the frequency of occurrence of these
functional morphemes is the relevant factor for their acquisition, as func-
tional morphemes with higher or lower frequency in the English and
French studies were different in their effects. That the German infants
seem to show a delay compared to the English and French infants may be
due to the much more complex article paradigm of German as compared
to both other languages. German has three gender classes and four case
categories, which are marked by six different forms of the singular definite
article (compared to two forms in French and one in English), and this
higher inventory must lead to a lower frequency of the individual forms of
the paradigm as compared to French or English. This facilitating effect of
high-frequency elements has not only been observed for functional mor-
phemes but also for other strings with a high frequency of occurrence in
the infants’ input, as for instance a child’s own name or the term used to
refer to the mother (for English: Bortfeld, Morgan, Golinkoff & Rathbun,
2005; for French: Mersad & Nazzi 2012). Beyond this, Japanese and Italian
infants as young as 8 months of age have been shown to be sensitive to the
typical order of high- and low-frequency elements in their respective
language and to use this information in the segmentation of speech
(Gervain, Nespor, Mazuka, Horie & Mehler 2008), again supporting an
early accessibility of frequency information.
The results for German, French and English uniformly show that chil-
dren learning these languages process and establish a representation of
functional elements from early on, which, however, initially seems not to
be fully phonologically specified. Due to their high frequency, functional
elements may well be accessible to infants’ processing and learning
mechanisms, which have been proven to be highly proficient in comput-
ing frequency distributions of sound patterns (Jusczyk, Luce & Charles-
Luce 1994, Mattys & Jusczyk 2001a, Maye, Werker & Gerken 2002, Onishi,
Chambers & Fisher 2002, Saffran, Aslin & Newport 1996). The crosslinguis-
tic comparison suggests that the acoustic salience of the realization of
functional elements in the speech stream does not make good predictions
about their acquisition. With respect to acoustics, English determiners
should be the least salient ones among the languages considered as they
are generally realized as unstressed syllables with schwa vowels. In German
determiners are unstressed as well, but the degree of vowel reduction is
generally lower than in English. In French the majority of determiners have
full vowels. If perceptual saliency as defined by these parameters deter-
mines the rate of acquisition, we would expect the English infants to be
the last in acquiring function words – an expectation that is contradicted by
the data. This raises the question as to whether stress is as crucial for infants’
speech processing as typically assumed (e.g. Bates & Goodman 1999,
Gleitman & Wanner 1982). So far, there is no empirical evidence supporting
the claim that infants have special problems in processing unstressed mate-
rial (Johnson 2005, Jusczyk & Thompson 1978). Adults’ disadvantages in the
processing of unstressed words might thus be the result of changing atten-
tional parameters (Cutler & Foss 1977, Cutler & Swinney 1987).
8.4.2 Bound functional morphemes

The processing and exploitation of information provided by bound func-
tional morphemes is a poorly studied field in early language acquisition.
Although there is growing evidence that young children can track non-
adjacent dependencies that involve bound functional morphemes dur-
ing their second year of life (for English: Santelmann & Jusczyk 1998,
Soderstrom, White, Conwell & Morgan 2007; for German: Höhle,
Schmitz, Santelmann & Weissenborn 2006; for Dutch: van Heugten &
Johnson 2010), research on the question of how the ability to process
these elements develops is scarce.
174 BARBARA HÖHLE
Blenn, Seidl and Höhle (2003) tested German 10-month-olds’ responsive-

ness to affixes of noun phrases by presenting them with noun phrases
with concordant morphology, i.e. every member of the phrase had the
same affix (e.g. diesen jungen Katzen ‘these young cats’) and the same phrases
involving non-concordant affixes (e.g. dieser jungen Katze ‘this young cat’). It
is important to note that both types of phrases are grammatical in German.
The infants listened longer to passages containing the phrases with con-
cordant markers as compared to passages with the non-concordant
phrases. A listening preference for the concordant phrases over the non-
concordant phrases was also obtained when the infants were initially
familiarized with passages containing these phrases and then were tested
for the isolated phrases (Pelzer & Höhle 2006). These results suggest that
the German infants are able to perceive these unstressed inflectional end-
ings and can discriminate them despite their high degree of segmental
overlap. As Pelzer and Höhle (2006) suggest, this sensitivity might help
infants to segment whole phrases marked concordantly out of continuous
speech. In German, the appearance of concordant phrases is restricted to
single instances of noun phrases depending on the gender of the noun, the
grammatical case and the phonological features of the noun itself. Thus,
concordant phrases in German are more the exception than the rule,
making the finding that the German children respond to this feature
even more intriguing. A question for further research would be how
children learning a language that makes more heavy use of concordant
morphology (e.g. Spanish or Italian) respond to it.
Another interesting question is whether infants process inflectional
endings as units that are partly independent from the stem they are
attached to. As inflectional endings typically occur with different stems,
distributional learning mechanisms should easily identify them as a unit
that is separable from the stem. Marquis and Shi (2012) present evidence
that 11-month-old French-learning babies have a representation of some
high-frequency inflectional endings of their language, and recognize the
relation between the non-inflected root form and its inflected variant.
They first familiarized infants with a non-inflected pseudoword. For the
test phase, the existing French suffix /e/ was added to the pseudoword and
it was presented together with a second, non-familiarized pseudoword
also carrying this inflectional ending. In this condition, the infants pre-
ferred to listen to the word with the familiar stem. However, in a control
condition, in which a segment not corresponding to a French inflectional
affix (/u/) was added to the familiarized pseudoword, no preference for
listening to this string was obtained.
Mintz (2013) used a similar technique with English-learning 15-month-
olds but reversed the order of presentation of the inflected and non-
inflected pseudowords. The infants were initially exposed to pseudowords
that either included the affix -ing or a non-existing pseudo-affix /dut/.
During the test phase, infants were exposed to the stems of both types of
pseudowords and showed longer listening times for the stems that had
carried the -ing ending during the familiarization. Taken together, the
results of these experiments show that infants at the beginning of their
second year of life start to respond differently to affixes present in their
language as compared to nonsense syllables. Furthermore, they seem to
begin to recognize a relation between a word that carries an inflectional
ending and the corresponding word stem without this inflectional ending,
which is essential for building up morphological paradigms. The fact that
functional morphemes typically occur at the edges of words – either as
prefixes or suffixes – may make them suitable to function as word bound-
ary cues as well.
8.5 Crosslinguistic issues in the syntactic categorization

of words
8.5.1 Categorizing words

Being able to segment words provides infants with one of the basic build-
ing blocks of grammar. However, to use words in an appropriate linguistic
context their syntactic properties have to be detected as well. One crucial
property that words share across languages is their fragmentation into
different classes according to their syntactic properties, with some degree
of variation with respect to the number of different categories among
languages. The question as to how the different syntactic categories are
established during language acquisition is a matter of intense debate.
Within nativist accounts it is assumed that the knowledge of the existence
of different syntactic categories is part of Universal Grammar and that the
child’s acquisition task consists in identifying instances of these categories
in the language they are learning (e.g. Pinker 1984 and Chapter 2). According
to Pinker’s semantic bootstrapping hypothesis, children are equipped with
universal linking rules between semantic properties and form classes. By
the use of these linking rules children assign a word referring to an object to
the class of nouns and a word referring to an action to the class of verbs. This
aids children in bootstrapping into a first lexicon involving syntactic cate-
gory information about the items included. This syntactic classification of
the first lexical items allows the child to perform an analysis of the distribu-
tional patterns the words typically occur in. These distributional patterns
replace the use of meaning-class relations as a more reliable cue to the
syntactic category membership of new words.
Accounts not sharing the assumption of an initial linguistic endowment
assume that syntactic categories emerge during the acquisition process by
mapping syntactic categories onto conceptual ones (Gentner 1982) or by
identifying similar features of initially syntactically non-categorized items
(Tomasello 2000b, and see Chapter 3). In other proposals, input cues like
phonological properties of the word forms themselves or distributional
176 BARBARA HÖHLE
information are considered as the basis for the construction of syntactic

categories (Maratsos & Chalkley 1980).
8.5.2 Phonetic and phonological cues to word categories

The discussion about a possible impact of phonetic or phonological infor-
mation on word categorization goes back to Gleitman and colleagues’
(Gleitman & Wanner 1982, Gleitman, Gleitman, Landau & Wanner 1988)
proposal that the correlation of stress and syntactic category, at least for
the closed–open class distinction, might be useful for the learner in dis-
covering the morphosyntactic distinctions typically associated with these
two categories.
Phonological correlations between open- and closed-class items corre-
sponding roughly to lexical and grammatical morphemes and their poten-
tial role in language acquisition have been the subject of some empirical
studies. Based on the observation that function words are typically more
minimal in their phonological form, Shi and colleagues (Shi, Morgan &
Allopenna 1998) investigated different features relevant for phonological
complexity vs minimality in English, Mandarin Chinese and Turkish
infant-directed speech. They found that in all three languages under inves-
tigation, the average lexical item had significantly more syllables, more
complex syllables, higher vowel durations and a higher relative amplitude
than the average functional item. Besides these features that hold for all
three languages, there were single cues only observed in individual lan-
guages, which depended on the specific phonological system of the respec-
tive language; for example, in Turkish lexical items were harmonic to the
preceding syllable in more cases than functional items, in Mandarin
Chinese more marked tones occurred in lexical than in functional items,
and in English a higher amount of vowel reduction was observed in func-
tional as compared to lexical items. Although comparing the means
yielded significant differences between the two classes, there was a high
amount of overlap with respect to each individual feature under study so
that none of the cues on its own had the power to allow a reliable assign-
ment of a given item to one of the two classes. However, computer simula-
tions with self-organizing neural networks showed that using these cues
simultaneously leads to a reliable assignment of items to the two gramma-
tical classes.
The data provided so far suggest that the input contains acoustic and
phonological cues that a learner might use for a rudimentary classification
into lexical and functional items. The question now is whether the learner
has the capacities to make use of these cues. Data by Shi, Werker and
Morgan (1999) suggest that this is the case. They found that English-learn-
ing infants made a categorical distinction between English lexical and
English function words that represented the typical features observed for
the two word classes.
There are indications that word forms may contain phonological cues
that allow a more finely graded categorization within these broad classes,
e.g. the categorization into nouns and verbs (Durieux & Gillis 2000, Kelly
1996). Acoustic differences in the pitch and duration patterns of homo-
phonous nouns and verbs have been observed by Shi and Moisan (2008) for
French and by Conwell and Morgan (2012) for English child-directed
speech. English-learning 13-month-olds were also able to discriminate
between homophonous nouns and verbs spliced out from child-directed
spontaneous speech, suggesting that they are sensitive to the relevant
acoustic differences (Conwell & Morgan 2012). But so far, there is no
clear empirical evidence that children in the age range considered here
use these cues to distinguish nouns and verbs as syntactically different
categories and use phonological cues to assign new words to one of these
categories.
8.5.3 Distributional cues to word category

Some more comprehensive research has looked at distributional informa-
tion as a cue to the syntactic categorization of words. From a linguistic
point of view distributional information should be the most reliable cue
for the syntactic categorization of word forms, because syntactic cate-
gories are established by words sharing the same distributional properties.
Based on the findings that children are sensitive to functional morphemes,
the proposal that functional morphemes provide important structural
information that children use to categorize content words is reasonable.
Functional morphemes can be seen as providing the structural frame of a
sentence with empty slots for the insertion of content words. Support for
structural frames is provided by Soderstrom and colleagues (Soderstrom
et al. 2007), who found that 16-month-old English learners notice the
misplacement of an inflectional ending but not the misplacement of a
non-inflected content word within a given sentence. This is in line with
findings by Shafer, Shucard, Shucard and Gerken (1998). They presented
10- and 11-month-old English learners with normal passages and with
passages in which some of the function words had been replaced by
nonsense syllables. Using the ERP technique they found differences in
the 11-month-olds’ brain responses for the normal and the modified pas-
sages, suggesting that these infants have some sensitivity to the distribu-
tion of elements with typical function word shape in speech. These
findings support the assumption that infants begin building a syntactic
structure based on function morphemes and their relationships. If this is
the case, these morphosyntactic structures provide crucial information on
the syntactic categories of the content elements appearing within them.
The first evidence for this scenario was presented by Brown (1957), who
reported that 3- to 5-year-olds’ interpretation of a new word is dependent
on its morphosyntactic environment, e.g. by relating a sib to a presented
178 BARBARA HÖHLE
object and sibbing to a presented action – a finding that has been verified by
a number of more recent studies with toddlers (Eyer et al. 2002, Gelman &
Markman 1985, Taylor & Gelman 1988) and with even younger children
(Bernal, Lidz, Millotte & Christophe 2007, Echols & Marti 2004, Katz,
Baker & Macnamara 1974, Waxman & Booth 2001, 2003).
A study with German learners suggests that the morphosyntactic envir-
onment not only helps the child to find a referent for a new word but that
the new word is assigned to a syntactic category with specific distribu-
tional properties. Höhle, Weissenborn, Kiefer, Schulz and Schmitz (2004)
presented 15-month-olds with noun phrases consisting of the German
indefinite article and a new pseudoword (ein pronk ‘a pronk’). After famil-
iarizing infants with these noun phrases the new word was presented in
sentences within another syntactic environment, either constituting
another frame for the noun use of the word (e.g. dieser pronk ‘this pronk’)
or constituting a frame for the verb use of the same new word (e.g. sie
pronk1 ‘she pronk’). The children showed a listening preference for the use
of the new word in the verb context, suggesting a novelty effect for the
ungrammatical structure. These results suggest that German learners
use the appearance of a determiner before an unknown word to assign
the new word to a syntactic category that we would call nouns. The fact
that they accept the use of the new word in environments that are lexically
different from but syntactically identical to the environment in which the
word had occurred before shows that children as young as 15 months have
some generalized knowledge about the syntactic features of at least some
syntactic classes and do not generally exploit syntactic knowledge in an
item-by-item fashion (Tomasello 2000b, Chapter 5).
Shi and Melançon (2010) found a similar asymmetry between noun and
verb categorization in French-learning 14-month-olds although they used
a task that was less challenging. Like Höhle et al. they familiarized infants
with sequences of either a French determiner and a pseudoword (e.g. ton
crale ‘your crale’) or of a French pronoun and a pseudoword (e.g. il mige ‘he
mige(s)’). In the test phase, the familiarized pseudoword was then com-
bined either with another determiner or with another pronoun either
from the same syntactic category (e.g. le crale ‘the crale’) or from the
other one (e.g. tu crale ‘you crale’). Only those infants who were familiar-
ized with the determiner/pseudoword combination responded to this
manipulation: they showed longer listening times to the pronoun/pseudo-
word combination than to the new determiner/pseudoword combination.
These findings mirror those obtained with the German children and
1
Note that the use of the new word as a verb form does not necessarily require the adding of an inﬂectional ending in
German. Furthermore, a replication of the experiment using a new pseudoword form that could also be an
inﬂected verb form (e.g. melt) yielded the same results. It is important for the interpretation of the results that a group of
infants familiarized with a pronoun context (er pronk – ‘he pronk’) and presented with the same sentences during
testing did not show the same effect.
suggest that noun categorization based on distributional information may

be easier than verb categorization.
However, Cauvet et al. (2014) did not find evidence for such a verb–noun
asymmetry in slightly older French 18-month-olds. Using a habituation
technique, they conditioned the infants to turn their head when a word
was presented which was preceded by either a determiner or a pronoun. In
the test phase, the target words were either presented in a syntactically
appropriate context in which the determiner was replaced by another
determiner or the pronoun by another pronoun, or they were presented
in a syntactically non-appropriate context in which the determiner had
been replaced by a pronoun and vice versa. Children were more successful
in detecting the nouns and the verbs in an appropriate context than in an
inappropriate context, suggesting that they can also make use of pronouns
as indicators that the following word will be a verb.
Mintz (2006) provided evidence that even younger English-learning
infants use distributional information to categorize new verbs. In contrast
to the studies reported so far, in which only bigram information (i.e. the
effect of one functional element on the following word) was provided,
Mintz tested the reliability of so-called frequent frames for assigning new
words to a syntactic category. Frequent frames are constituted by non-
adjacent word pairs with a variably filled one-word slot between them that
occur with a high frequency in the child’s input (Mintz 2003), such as, to . . .
it. The elements constituting the frequent frames are not necessarily
function words. Mintz argues that the frequency of co-occurrence of the
words constituting the frame by itself makes it likely that the existence of
the frame reflects a systematic aspect of the language and is not a product
of chance. This in turn suggests that the words occurring within this frame
share systematic properties like the syntactic category. From the analyses
of several corpora of child-directed speech, Mintz (2003) could show that
child-directed speech contains frequent frames of the above type and that
in fact the different words occurring in these frames had a high degree of
overlap with respect to their syntactic category. Interestingly, most of the
frequent frames observed in these corpora were frames for verbs.
Mintz (2006) tested whether 12-month-old English-learning children
would be able to use the information given by frequent frames for a
syntactic categorization of the words occurring within the frames.
Similarly to the method used by Höhle et al. (2004) and Shi and Melançon
(2010), infants were familiarized with new words within a context provid-
ing a frequent frame either for a noun or for a verb and then were tested
with the same words either in a different frame for the familiarized
category or in a different frame indicating another syntactic category for
the enclosed word. The English infants showed a listening preference
for the presentation of the new words in a frame indicating a different
syntactic category from the familiarized one. These results show that at
the end of the first year English-learning children can use distributional
180 BARBARA HÖHLE
information to categorize new verbs. It is an open question whether the

fact that this ability has been shown for younger English infants is due to
the use of frequent frames as more reliable markers of the syntactic
category of verbs in this study or whether the ability to categorize verbs
in general emerges earlier in English-learning children compared to lear-
ners of German or French due to language-specific properties.
This raises the question of whether the concept of frequent frames can
be applied to other languages that have, for example, a higher degree of
word-order variation or a more complex morphological system than
English. As frequent frames are defined by pairs of specific word forms,
word-order variation or a rich inflectional system might reduce their
number considerably. The existence of frequent frames has meanwhile
been attested across a wider range of languages (French: Chemla, Mintz,
Bernal & Christophe 2009; Spanish: Weisleder & Waxman 2010), although
in some languages they seem to provide less reliable information for the
categorization of words (German: Stumper, Bannard, Lieven & Tomasello
2011; Dutch: Erkelens 2009). However, it is possible that in languages with
more regular affixation processes like Turkish, affixes also should be
considered as parts of frequent frames (Wang, Höhle, Ketrez, Küntay &
Mintz 2011).
Young children can also exploit distributional information to make
subcategorical distinctions within broader word classes, for example,
those involving gender categories. Cyr and Shi (2012) found evidence
that French-learning 20-month-olds can generalize across different deter-
miners marked for gender: they familiarized infants with a pseudo-noun
that was preceded by a gender-marked indefinite article (un or une). During
the test phase, the children were presented with the same pseudo-nouns
but now preceded by the definite article (le or la), either gender congruent
with the familiarization or not. Infants preferred the gender congruent
condition suggesting that they have gained some knowledge about the
French gender categories, at least about the fact that un and le (the mascu-
line forms) share distributional properties that are different from those
shared by the feminine forms une and la and thus that two subcategories
exist within the determiners.
These findings converge with findings from Gerken, Wilson and Lewis
(2005), who tested the use of Russian inflectional endings as markers for
gender categories with 16-month-old American infants. During a familiar-
ization phase, the infants were presented with pairs of Russian masculine
or feminine words which appeared with two different gender-specific case
markers. During the test phase, words were presented that had only been
presented with one of the two possible endings during the familiarization
phase. This word carried either a gender congruent case marking or a
gender incongruent case marking. The infants showed longer listening
times to the word with an incongruent case marker – but only when
additional information in the word stem indicated the gender category.
However, the results suggest that rather young infants have the capacity
to reconstruct morphological paradigms based on grammatical gender
categories even if this category is not relevant in the language they are
learning.
8.6 Some conclusions
Our overview has shown that languages provide a number of cues that may
support infants’ initial steps into the segmentation and the syntactic
categorization of words in their ambient language. Furthermore, infants
seem to become attuned to the specific properties of their language rather
fast and start to exploit those cues that are informative in their language
quickly. In a nutshell, there might be two kinds of information that are
especially relevant for the early steps of young children into language
acquisition, namely rhythmic information and distributional information,
both of which operate on different linguistic levels including the pho-
neme, the syllable and the word. But, again, evidence for infants’ exploita-
tion of this kind of information stems from experiments with highly
controlled stimuli and it is unclear whether they can cope with the varia-
bility that characterizes natural speech (Johnson & Tyler 2010).
Several studies have raised the question of whether there is a dominance
relation between rhythmical and segmental distributional cues in that one
type of cue is used earlier or weighted over the other by young children
and whether there is a developmental change concerning the weighting
of these cues across age (Johnson & Jusczyk 2001, Mattys et al. 1999,
Thiessen & Saffran 2003). The use of distributional cues seems to be less
dependent on language-specific properties, but the evidence for the exploi-
tation of distributional cues comes mostly from highly controlled and
restricted experimental conditions; hence, their suitability for processing
the variability that is provided by the exposure to natural language has
been questioned (Johnson & Tyler 2010). So far, the research on the exploi-
tation of these cues in solving basic problems in language acquisition has
mostly focused on a single cue or a combination of a few cues for segmen-
tation and categorization. This is an appropriate approach if the question
as to whether infants can make use of specific kinds of information is in
the foreground. However, exposure to natural language in situations of
social interaction provides the infant with a whole range of different cues
in parallel, such as intonational cues for boundaries of larger prosodic
units that coincide with word boundaries (Gout, Christophe & Morgan
2004, Shukla, Nespor & Mehler 2007, Shukla, White & Aslin 2011) or
specific characteristics of infant-directed speech (Johnson, Seidl & Tyler
2014, Thiessen, Hill & Saffran 2005). Accordingly, research on how these
different sources of information interact in their exploitation by the child
is a further challenge for future research.
182 BARBARA HÖHLE
Furthermore, given the different types of cues that languages offer it

seems plausible that there are also differences in cue weighting across
languages. For example, it is possible that French infants can rely more
heavily on transitional probabilities across syllables since French as a
syllable-timed language is easier to segment into syllables than stress-
timed languages, for which a segmentation strategy based on rhythmic
properties may be more appropriate. Or, to mention another example,
French has an obligatory use of determiners while in German not every
noun phrase must include a determiner and due to stronger gender and
case differentiation German has a higher number of different determiner
forms. As we have seen, French infants can recognize determiners in their
language at a very early age and seem to use determiners as a segmentation
cue at a much younger age than German infants. This suggests that even in
languages with an overall high number of structural similarities, differ-
ences in the trajectory of early language acquisition can be observed. In
this case differences may only concern the timing as both languages share
the basic information that infants use. However, the available evidence on
how children start speech segmentation and categorization as well as the
changes which these processes may undergo is still based on a handful of
languages that only represent a minimal proportion of the approximately
6,000 different languages across the world (Dryer & Haspelmath 2013).
Taking into consideration the great variation in the phonological and
the morphosyntactic systems across languages, our understanding of the
interaction between the learning mechanisms that are available to the
child and their optimal adoption given the specific structure of the lan-
guage to be learned remains fragmentary. Much more comparative acqui-
sition research on a broader variety of languages – specifically focusing on
typologically very different groups like tone languages or polysynthetic
languages – using the same methods and comparable materials is needed.
Due to the increased technical requirements of the research with very
young children this is still a challenging enterprise, but initial steps in
this direction haven been taken (Friederici, Friedrich & Christophe 2007,
Gervain et al. 2008, Höhle et al. 2009, Mattock & Burnham 2006, Skoruppa et
al. 2009) and should be strengthened in the future.
Suggestions for further reading
Jusczyk, P. W. (1997). The Discovery of Spoken Language. Cambridge, MA: MIT

Press.
Nazzi, T., & Ramus, F. (2003). Perception and acquisition of linguistic
rhythm by infants. Speech Communication, 41, 233–43.
Shi, R. (2014). Functional morphemes and early language acquisition. Child
Development Perspectives, 8, 6–11.

Cambridge Books Online

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cambridge Books Online

Uploaded by

Copyright:

Available Formats

Cambridge Books Online

The Cambridge Handbook of Child Language

Edited by Edith L. Bavin, Letitia R. Naigles

Book DOI: http://dx.doi.org/10.1017/CBO9781316095829

Online ISBN: 9781316095829

Hardback ISBN: 9781107087323

Paperback ISBN: 9781107455504

8 - Crosslinguistic perspectives on segmentation and categorization in

early language acquisition pp. 159-182

Chapter DOI: http://dx.doi.org/10.1017/CBO9781316095829.008

Cambridge University Press

variations. Even though the crosslinguistic variation can be described

8.2 Some methodological remarks

Because we are examining an early phase of language acquisition compris-

8.3 Crosslinguistic issues in word segmentation

8.3.1 Rhythmical typology and rhythmical sensitivity

reflect the differences between languages with respect to their typical

or in segmentally less variable strings than their Spanish-learning age

8.3.2 Using rhythm to segment speech

syllable – is a reliable cue for an initial word boundary for a reasonable

8-month-old infants learning Canadian French can successfully segment

studies that tested the segmentation of monosyllabic words. According to

8.3.3 Further phonological cues to word boundaries

(2008) first tested Turkish-learning 6-month-olds’ sensitivity to this feature

8.4 Functional morphemes and their role in segmenting

In many languages, functional morphemes are essential for indicating the

8.4.1 Free functional morphemes

always included the critical determiner in initial position, Höhle and

8.4.2 Bound functional morphemes

Blenn, Seidl and Höhle (2003) tested German 10-month-olds’ responsive-

8.5 Crosslinguistic issues in the syntactic categorization

8.5.1 Categorizing words

information are considered as the basis for the construction of syntactic

8.5.2 Phonetic and phonological cues to word categories

8.5.3 Distributional cues to word category

suggest that noun categorization based on distributional information may

information to categorize new verbs. It is an open question whether the

8.6 Some conclusions

Furthermore, given the different types of cues that languages offer it

Suggestions for further reading

Jusczyk, P. W. (1997). The Discovery of Spoken Language. Cambridge, MA: MIT

You might also like