In the present study, we first started with very primitive analyses of production units for continuous speech “Chunk” and “IPU”. Although Chunk length did not prove to be as effective as IPU length in the subject-based LME models in predicting overall speech intelligibility, both of them differed among the HA, CI and NH children. The results are encouraging, as the text-based Chunks and the speech-based IPUs are identified fully automatically and they to a certain degree reflect the ability of children in terms of lexical complexity. Compared with MLU, the measures we proposed are less labor intensive. Chunks or IPUs may be considered as supplementary indicators in addition to MLU. Furthermore, word-level acoustic-phonetic measures have proven to be useful for distinguishing group differences among the NH, HA and CI children. In the era of big data analysis and computational modeling, automated, objective measures that are motivated from linguistically interpretable properties can empirically contribute to developing effective assessment tools and learning interfaces, e.g., lexical tone acquisition. Tone production in Mandarin Chinese involves important phonological properties that often lead to difficulties in speech acquisition of hard of hearing children (Han et al. 2007). Utilizing computational modeling of fundamental frequency values, quantitative representation of tone contours can substantially enhance qualitative analysis of tone production. Diagnoses, targeted interventions and training programs that are normally conducted by professional experts may be strengthened in a similar way by applying approaches from the fields of natural language processing and automatic speech processing.