In this study, we aim to automatically score the spoken responses from an international English assessment targeted to non-native English-speaking children aged 8 years and above.In contrast to most previous studies focusing on scoring of adult non-native English speech, we explored automated scoring of child language assessment.We developed automated scoring models based on a large set of features covering delivery (pronunciation and fluency), language use (grammar and vocabulary), and topic development (coherence).In particular, in order to assess the level of grammatical development, we used a child language metric that measures syntactic proficiency in emerging language in children.Due to acoustic and linguistic differences between child and adult speech, the automated speech recognition (ASR) of child speech has been a challenging task.This problem may increase difficulty of automated scoring.In order to investigate the impact of ASR errors on automated scores, we compared scoring models based on features from ASR transcriptions with ones based on human transcriptions.Our results show that there is potential for the automatic scoring of spoken non-native child language.The best performing model based on ASR transcriptions achieved a correlation of 0.86 with human-rated scores.

The index of productive syntax (IPSyn; Scarborough (Applied Psycholinguistics 11:1–22, 1990) is a measure of syntactic development in child language that has been used in research and clinical settings to investigate the grammatical development of various groups of children. However, IPSyn is mostly calculated manually, which is an extremely laborious process. In this article, we describe the AC-IPSyn system, which automatically calculates the IPSyn score for child language transcripts using natural language processing techniques. Our results show that the AC-IPSyn system performs at levels comparable to scores computed manually. The AC-IPSyn system can be downloaded from www.hlt.utdallas.edu/~nisa/ipsyn.html.

Child language narratives are used for language analysis, measurement of language development, and the detection of language impairment. In this paper, we explore the use of Latent Dirichlet Allocation (LDA) for detecting topics from narratives, and use the topics derived from LDA in two classification tasks: automatic prediction of coherence and language impairment. Our experiments show LDA is useful for detecting the topics that correspond to the narrative structure. We also observed improved performance for the automatic prediction of coherence and language impairment when we use features derived from the topic words provided by LDA.

Language is an important part in a human being's life. Identification of language deficiencies such as the use of certain syntactic constructs and words in children with Language Impairment (LI) will allow clinicians to create intervention programs that focus on these deficiencies. Most measures of language development and LI are computed manually which is time consuming. Automatic computation of language development measures will allow for a quicker and more extensive study of language samples. This dissertation presents the use of Natural Language Processing (NLP) techniques for the study of child language analysis. We first explore the use of NLP in child language analysis from a syntactic perspective. We present AC-IPSyn, a system that automates the computation of the Index of Productive Syntax (IPSyn) (Scarborough, 1990), a metric that measures syntactic complexity in child language. AC-IPSyn performs at levels comparable to human scoring. We use syntactic features based on IPSyn in the automatic prediction of LI. We then conduct a study of grammatical errors in child language transcripts. We annotate child language transcripts for verb related grammatical errors and present an automatic grammar error detection system. We then explore the use of NLP from the semantic aspect of language. The ability to produce coherent language is a hallmark of language development. We conduct a study of coherence using child language narratives. We annotate story telling narratives for coherence, narrative structure, and narrative quality elements. We then use coherence and narrative related features in the automatic prediction of coherence and LI. To avoid the labor intensive process of manual annotation of narrative structure, we explore automatic identification of topics from narratives. For this purpose, we use Latent Dirichlet Allocation (LDA), a topic modeling method. Our results show LDA is useful for detecting the topics that correspond to the narrative structure. The usage of coherence, narrative, and topic related features were useful in the automatic prediction of coherence and LI. Finally, we generalize our work to second language assessment. We present an automatic scoring system for child speech on an assessment test for children studying English as a foreign language. We explore the use of speech, grammar, and coherence features. Our results show that automatic scoring of child language is promising.

Language impairment (LI) in children is pervasive in all walks of life. Automatic prediction of LI is useful as a first pass for speech language pathologists in identifying prospective children with LI. Previous work in the automatic prediction of LI has explored various features, mostly shallow and surface level features. In this paper, we evaluate deeper Natural Language Processing (NLP) features such as syntactic, semantic and entity grid model features, along with narrative structure and quality features in the prediction of LI using child language transcripts. Our experiments show that narrative structure and quality features along with a combination of other features are helpful in the prediction of LI in storytelling narratives.

Coherence is an important aspect of language ability. In this study, we analyze and annotate child language samples of story retell sessions for coherence and presence of narrative structure and narrative quality constructs. We use these constructs as features and use existing Natural Language Processing (NLP) techniques to build models that automatically predict coherence and language impairment in narratives. Our feature analysis results give us an insight into some of the important narrative quality features such as the use of cognitive inferences and social engagement devices. Our study shows that modeling of coherence in the context of language development in children is promising.

Language sample analysis is an important technique used in measuring language development. At present, measures of grammatical complexity such as the Index of Productive Syntax (Scarborough, 1990) are used to measure language development in early childhood. Although these measures depict the overall competence in the usage of language, they do not provide for an analysis of the grammatical mistakes made by the child. In this paper, we explore the use of existing Natural Language Processing (NLP) techniques to provide an insight into the processing of child language transcripts and challenges in automatic grammar checking. We explore the automatic detection of 6 types of verb related grammatical errors. We compare rule based systems to statistical systems and investigate the use of different features. We found the statistical systems performed better than the rule based systems for most of the error categories.

Objectives: This paper explores the use of an automated method for analyzing narratives of monolingual English speaking children to accurately predict the presence or absence of a language impairment. The goal is to exploit corpus-based approaches inspired by the fields of natural language processing and machine learning. Methods and materials: We extract a large variety of features from language samples and use them to train language models and well known machine learning algorithms as the underlying predictors. The methods are evaluated on two different datasets and three language tasks. One dataset contains samples of two spontaneous narrative tasks performed by 118 children with an average age of 13 years and a second dataset contains play sessions from over 600 younger children with an average age of 6 years. Results: We compare results against a cut off baseline method and show that our results are far superior, reaching F-measures of over 85% in two of the three language tasks, and 48% in the third one. Conclusions: The different experiments we present here show that corpus based approaches can yield good prediction results in the problem of language impairment detection. These findings warrant further exploration of natural language processing techniques in the field of communication disorders. Moreover, the proposed framework can be easily adapted to analyze samples in languages other than English since most of the features are language independent or can be customized with little effort.

This paper describes a technique for automatically tagging political blog posts using SVM's and named entity recognition. We compare the quality of the tags detected by this approach to earlier approaches in other domains, observing effects from the political domain and benefits from NLP techniques complementary to the core SVM method.