We examined word use of our subjects by CKIP POS and UD word categories. As shown in Figure 3a, with the ten most frequently used POS marked on the x-axis, patterns in terms of CKIP POS are quite similar across the three subject groups. UD word categories in Figure 3b show the same tendency. Regarding lexical use, hard of hearing children who have received regular training (in our case, AVT intervention) performed narratives in a comparable way as did the hearing subjects. As shown in the results, the NH children show lexical preferences for pronouns and fillers. Pronouns were produced in 7.1%, 4.6%, and 3.8% of the overall data produced by the NH, HA, and CI children. The NH children tended to use pronouns to denote known objects (given information) more frequently than did the HA and CI children. Particularly interesting is the use of fillers, which were produced at rates of 2.3%, 0.6%, and 0.9% in the NH, HA and CI children. Table 5 lists all the filler tokens produced by our subjects. For a complete list of fillers in Chinese conversation, please refer to Tseng (2013). The bilabial variant mhm and the variant with a prolonged nasal coda mhmm are clearly preferred by the NH children. Although fillers are usually regarded as peripheral phenomenon, the use of fillers in communicative, spontaneous speech is particularly indicative of socio-pragmatic skills, if used properly. Acquiring speech ability may accordingly include the ability (or strategy) of adequately using fillers that allow more time for the speakers in interactive contexts. To our surprise, our analysis of word use reveals this difference between hearing children and children with hearing impairment.