Discovering a word’s part of speech is an essential step in acquiring the grammar of a language. In this thesis we examine a variety of computational Bayesian models that use linguistic input available to children, in the form of transcribed child directed speech, to learn part of speech categories. Part of speech categories are characterised by contextual (distributional/syntactic) and word-internal (morphological) similarity. In this thesis, we assume language learners will be aware of these types of cues, and investigate exactly how they can make use of them. Firstly, we enrich the context of a standard model (the Bayesian Hidden Markov Model) by adding sentence type to the wider distributional context.We show that children are exposed to a much more diverse set of sentence types than evident in standard corpora used for NLP tasks, and previous work suggests that they are aware of the differences between sentence type as signalled by prosody and pragmatics. Sentence type affects local context distributions, and as such can be informative when relying on local context for categorisation. Adding sentence types to the model improves performance, depending on how it is integrated into our models. We discuss how to incorporate novel features into the model structure we use in a flexible manner, and present a second model type that learns to use sentence type as a distinguishing cue only when it is informative. Secondly, we add a model of morphological segmentation to the part of speech categorisation model, in order to model joint learning of syntactic categories and morphology. These two tasks are closely linked: categorising words into syntactic categories is aided by morphological information, and finding morphological patterns in words is aided by knowing the syntactic categories of those words. In our joint model, we find improved performance vis-a-vis single-task baselines, but the nature of the improvement depends on the morphological typology of the language being modelled. This is the first token-based joint model of unsupervised morphology and part of speech category learning of which we are aware.

It has long been argued that prosodic cues may facilitate syntax acquisition (e.g., Morgan, 1986). Previous studies have shown that infants are sensitive to violations of typical correlations between clause-final prosodic cues (Hirsh-Pasek et al., 1987) and that prosody facilitates memory for strings of words (Soderstrom et al., 2005). This dissertation broaches the question of whether children can use this information in syntax acquisition by asking if learners can use the prosodic correlates of clauses to locate syntactic constituents. One property of certain syntactic constituents in natural languages is that they can move, so learning of constituency was inferred if participants treated prosodically-grouped words as cohesive, moveable chunks. In Experiment 1, 19-month-olds were familiarized with sentences from an artificial grammar with either 1-clause or 2-clause prosody. The infants from the 2-clause group later recognized the prosodically-marked clauses when they had moved to a new position in the sentence and had a new acoustic contour. Adults in Experiment 2 showed similar learning, although their judgments also rely on recognition of perceptually-salient words at prosodic boundaries. Subsequent experiments explored the mechanisms underlying this prosodic bootstrapping by testing Japanese-acquiring infants on English-based stimuli (Experiment 3) and English-acquiring infants on Japanese-based stimuli (Experiment 4). Infants were able to locate constituent-like groups of words with both native and non-native prosody, suggesting that the acoustic correlates of prosody are sufficiently robust across languages that they can be used in early syntax acquisition without extensive exposure to language-specific prosodic features. On the other hand, adults (Experiment 5) are less flexible, and are only able to use prosody consistent with their native language, suggesting an age- or experience-related tuning of the prosodic perceptual mechanism. This dissertation supports prosody as an important cue that allows infants and young children to break into syntax even before they understand many words, and helps explain the rapid rate of syntax acquisition.

Sprachverständnisuntersuchungen sind sowohl für die Forschung zum Spracherwerb und zu dessen Störungen als auch für die Diagnostik in der klinischen Praxis von essentieller Bedeutung. Zwei der verbreitetesten Methoden zur Erfassung des Verständnisses von syntaktischen Strukturen sind das Ausagieren von Sätzen (AS) und die Satz-Bild-Zuordnung (SBZ). Beide Methoden sind mit unterschiedlichen Anforderungen an die Probanden sowie mit spezifischen Vor- und Nachteilen verbunden (vgl. z.B. Goodluck 1996, Gerken & Shady 1996), woraus sich die Frage ergibt, inwiefern die Wahl der Methode das Testergebnis und die damit verbundenen Schlussfolgerungen beeinflusst. In der hier beschriebenen empirischen Untersuchung wurden die AS- und die SBZ-Methode anhand des Verstehens semantisch reversibler Sätze mit Objektvoranstellung direkt miteinander verglichen: Jeweils 24 monolingual mit Deutsch als Muttersprache aufwachsenden, sprachunauffälligen Kindern im Alter von 3;0 bis 3;5 bzw. 5;0 bis 5;5 Jahren wurden insgesamt zwölf Items, von denen neun aus semantisch reversiblen Sätzen mit Objektvoranstellung und drei aus entsprechenden subjektinitialen Kontrastsätzen bestanden, sowohl mittels der AS-Methode als auch mittels der SBZ-Methode präsentiert. Zum Ausagieren der Testsätze (AS-Test) wurden Spielzeugfiguren verwendet, für die Satz-Bild-Zuordnung (SBZ-Test) wurden pro Item ein Ziel- und zwei Ablenkerbilder erstellt. Die kindlichen Reaktionen wurden nach syntaktischen Kriterien als korrekt bzw. inkorrekt bewertet, Abweichungen wurden hinsichtlich der Fehlerart klassifiziert. Zusätzlich wurde bei den einzelnen Kindern der Erwerbsstand der untersuchten Struktur beurteilt und es wurden, sofern möglich, die eingesetzten Sprachverständnisstrategien ermittelt. Der Vergleich der Methoden ergab für die einzelnen Altersgruppen sehr unterschiedliche Befunde: Bei den 3;0- bis 3;5-jährigen Kindern zeigten sich im SBZ-Test teilweise starke Rateeffekte, die zu verfälschten Ergebnissen und kaum interpretierbaren Reaktionsmustern führten, während dies im AS-Test nicht der Fall war. Für diese Altersgruppe ist demnach die AS-Methode als die geeignetere anzusehen. Bei den 5;0- bis 5;5-jährigen Kindern konnte indes keine derartige Diskrepanz zwischen den beiden Tests festgestellt werden. Vielmehr waren die Leistungen hier insgesamt vergleichbar, so dass für diese Altersgruppe die AS- und die SBZ-Methode als für die Untersuchung des Syntaxverständnisses gleichwertig gelten können. Darüber hinaus bestätigte die Untersuchung viele der in der Literatur beschriebenen Vor- und Nachteile der beiden Methoden, darunter der geringere Materialbeschaffungsaufwand bei der AS-Methode, die größere Durchführungspraktikabilität der SBZ-Methode, die höhere Attraktivität und das geringere Risiko für Antwort-biases bei der AS-Methode, die leichtere Bewertbarkeit der kindlichen Reaktionen bei der SBZ-Methode. Im Vergleich der Altersgruppen zeigte sich, dass die 5;0- bis 5;5-jährigen Kinder Testsätze mit Objektvoranstellung deutlich besser verstanden als die 3;0- bis 3;5-jährigen: Die meisten der jüngeren Kinder hatten die untersuchte Struktur offenbar noch nicht erworben – sie interpretierten die Sätze mit Objektvoranstellung stattdessen am häufigsten nach der Wortreihenfolgestrategie, verstanden sie also wie subjektinitiale Strukturen. Von den 5;0- bis 5;5-jährigen Kindern dagegen hatten 75% die untersuchte Struktur rezeptiv erworben, interpretierten die Testsätze also primär korrekt nach der grammatischen Strategie. Aus diesen Ergebnissen wurde geschlossen, dass die Struktur semantisch reversibler Sätze mit Objektvoranstellung frühestens ab einem Alter von 5;0 bis 5;5 Jahren als in der ungestörten Sprachentwicklung rezeptiv erworben gelten kann. Da sich jedoch in beiden Altersgruppen trotz allem eine erhebliche interindividuelle Variation hinsichtlich der Verständnisfähigkeiten zeigte, muss insgesamt von einem relativ langen Zeitfenster für den ungestörten Erwerb dieser Struktur ausgegangen werden. / The assessment of children’s receptive language abilities plays an essential part in both experimental research on language acquisition and clinical diagnostics of language acquisition impairments. Two of the most prevalent methods especially used for assessing syntax comprehension are act-out and picture selection. The two methods do not only differ in their requirements on the proband/patient, but also have specific advantages and disadvantages concerning the development, administration, and interpretation of experimental or diagnostic tests (cf. Goodluck 1996, Gerken & Shady 1996). This leads to the question of how far the adopted method influences the result of the test and the conclusions drawn from it. In the study reported here, act-out and picture selection were, by means of a comprehension task using semantically reversible sentences with object fronting, directly compared to each other. Subjects were 48 German-speaking children with normal language development: 24 aged 3;0 to 3;5 years and another 24 aged 5;0 to 5;5 years. The children were presented with twelve stimuli, nine consisting of semantically reversible sentences with object fronting and three of correspondent subject initial constructions, using both an act-out procedure with toy props and a picture selection procedure with three choices per sentence. The child’s responses were scored correct or incorrect according to syntactic criteria, with incorrect responses categorised by type. In addition, the child’s level of acquisition of the structure in focus and, when possible, the comprehension strategies used by the child were determined. The comparison of the two methods yielded very different results for the individual age groups: A considerable proportion of the 3;0 to 3;5-year-olds displayed strong effects of guessing with the picture selection procedure, which led to inaccurate results and response patterns that were hardly interpretable, whereas this was not the case with the act-out procedure. According to these findings then, act-out is to be considered the more suitable method for children of this age. However with the 5;0 to 5;5-year-olds, no such discrepancy between the two methods was found, their results in both tests being comparable. Thus, for children aged five years or older, picture selection and act-out can be regarded as equivalent methods for the assessment of syntax comprehension. In addition to these findings, many of the advantages and disadvantages of the two methods described in the literature were confirmed in the study, such as the lower material provision costs with the act-out procedure, the higher administration practicability of the picture selection method, the greater appeal for children and the lower risk for response-biases with act-out tasks, and with picture selection, the easier scoring of the child’s responses . Comparing the two age groups, the 5;0 to 5;5-year-olds showed considerably better comprehension of the semantically reversible sentences with object fronting than the 3;0 to 3;5-year-olds: The majority of the younger children obviously had not yet acquired the structure in question and instead, mostly relying on the so called word order strategy, misinterpreted the sentences with object fronting as subject-initial structures. In contrast, 75% of the older children had receptively acquired the syntax of object fronting, since they interpreted most of the test sentences correctly according to a grammatical strategy. From these results, it was concluded that the structure of semantically reversible sentences with object fronting can be regarded as receptively acquired in normal language development at the earliest at age 5;0 to 5;5 years. However, since there was considerable interindividual variation in the comprehension abilities in both age groups, the time frame for the acquisition of this structure must be assumed to be relatively large.