Effects of Phonotactics in Patients with Cochlear Implants

Probabilistic phonotactics refers to the frequency with which segments and sequences of segments occur in syllables and words. Knowledge of phonotactics has been shown to be an important source of information in segmenting and recognizing speech in normal hearing listeners. A post-perceptual task (nonword rating) and two on-line tasks (an auditory same-different and an auditory lexical decision task) were used in the present set of experiments to examine the use of phonotactic information by postlingually deafened adults who have received a cochlear implant. The results of all three experiments showed that both normal-hearing and hearing-impaired listeners are sensitive to differences in phonotactic information to varying degrees. Furthermore, cochlear implant patients with better word recognition abilities (as measured by the NU6) tended to be more sensitive to phonotactic information than cochlear implant patients with poorer word recognition abilities. The implications of these results for outcome assessments and clinical interventions are discussed. Phonotactic information refers to the sequential arrangement of phonetic segments in morphemes, syllables, and words (Crystal, 1980). Sounds and sequences of sound that are found in a given language are said to be legal within that language, whereas sounds and sequences of sound that are not found in a given language are said to be illegal within that language. Awareness of the sounds that are legal in one’s native language occurs very early in life. For example, Jusczyk, Frederici, Wessels, Svenkerud, and Jusczyk (1993) showed that Dutch and American children as young as nine months of age listen longer to lists of words with patterns of segments and sequences allowed in their native language than to lists of words with patterns from the other language. These results show that listeners are sensitive early in life to the sounds and sequences of sound that are legal in their native language. Although phonotactic information is often described as a set of rules—or a “phonological syntax” (Malmkaer, 1991)—specifying the sequences of segments that are legal or illegal in a language, recent work has explored the probabilistic nature of phonotactic constraints (Kessler & Treiman, 1997; Treiman, Kessler, Knewasser, Tincoff, & Bowman, 1996; Vitevitch & Luce, 1998, 1999; Vitevitch, Luce, CharlesLuce, & Kemmerer, 1997). That is, rather than using stimuli that contained either legal or illegal sequences, these researchers created stimuli that were completely legal in a given language, but that varied in how common the segments and sequences were in that language. Jusczyk, Luce, and CharlesLuce (1994) demonstrated that nine-month old infants are also sensitive to the probabilities of sound patterns within their native language. Using the same procedure as Jusczyk et al. (1993), Jusczyk, Luce, and Charles-Luce (1994) found that American infants listened longer to lists of nonwords that contained high probability segments and sequences of segments than to lists of nonwords that contained low probability segments and sequences in English. These results suggest that sensitivity to probabilistic phonotactic information may also develop early in life and may be important for the processing of spoken language later in life. For example, sensitivity to the phonotactic probabilities of the ambient language may assist children in acquiring and building a lexicon. Computational (e.g., Brent & Cartwright, 1996; Cairns, Shillcock, Chater, & Levy, 1997) and experimental investigations (e.g., Mattys, Jusczyk, Luce & Morgan, 1999; Saffran, Newport, & Aslin, 1996) suggest that phonotactic information may play a role in the segmentation of words from continuous speech. Sensitivity to the patterns of segments that occur only within words (such as /tl/), or only at the edges of words (e.g., // does not occur in the initial portion of EFFECTS OF PHONOTACTICS IN PATIENTS WITH COCHLEAR IMPLANTS 191 English words) may allow a child to identify the beginning and endings of words. With the beginning and ending of a word identified, the child can isolate an individual word from the continuous stream of speech and begin acquiring a lexicon. Other research suggests that older children may also use phonotactic information to add new words to the lexicon (e.g., Gathercole, Willis, Emslie, & Baddeley, 1991; Gupta & MacWhinney, 1997; Storkel & Rogers, 2000). Thus, phonotactic probabilities in language are valuable sources of information early in life for processing spoken language. Phonotactic information is not only used in the acquisition of language. Adults are also sensitive to phonotactic information and may use it to process spoken language. For example, Vitevitch, Luce, Charles-Luce, and Kemmerer (1997; see also Messer, 1967) created bisyllabic nonword stimuli containing segments and sequences of segments that were completely legal in English, but varied in how common they were in English. Stimuli comprised of segments and sequences of segments that occur frequently in English, such as /kikrig/, are said to have high phonotactic probability. Stimuli comprised of segments and sequences of segments that occur less frequently in English, such as /j^^t/, are said to have low phonotactic probability. The researchers asked participants to rate how “good” each item would be if it were a real word in English. Their results showed that participants’ subjective ratings of the spoken nonwords followed the objective measure of phonotactic probability: Nonwords with high probability patterns were rated as being more word-like than low probability patterns. These results suggest that adults are sensitive to fine-grained probabilistic phonotactic information within their native language, and can access and use this information in tasks requiring explicit judgment about nonword patterns. Vitevitch et al. (1997) also asked another group of participants to repeat the same nonwords presented auditorily. An analysis of the response latencies showed that nonwords with high probability patterns were repeated more quickly than nonwords with low probability patterns, suggesting that probabilistic phonotactic information may play a role in spoken word recognition in normal hearing listeners (see also Vitevitch & Luce, 1998, 1999; Vitevitch, Luce, Pisoni, & Auer, 1999). That is, phonotactic information may be one of several sources of information, such as word frequency (e.g., Savin, 1963; Solomon & Postman, 1952) or the stress pattern of a word (e.g., Cutler & Norris, 1988) that normal hearing listeners use to understand spoken language. In the present study, we were interested in determining whether a group of post-lingually deafened adults who have subsequently received a cochlear implant also make use of phonotactic probabilities to understand spoken words. Doyle et al. (1995), for example, reported that cochlear implant users have difficulty distinguishing among segments varying in manner of articulation, voicing, and place of articulation. Given the difficulty in discriminating fine phonetic details in speech, cochlear implant users may no longer consistently rely on information or representations related to segments or sequences of segments to process spoken words. Post-lingually deafened adults who used a cochlear implant for at least one year participated in the present set of experiments. Our goal was to determine if these patients are able to make use of information about phonotactic probabilities and whether these cognitive processing strategies help cochlear implant users recognize isolated spoken words. The post-lingually deafened adults who participated in this set of experiments were all patients who had acquired language with normal hearing. Later in life these individuals became profoundly deafened through trauma or disease and had subsequently received and used a cochlear implant for at least a year. A cochlear implant is a sensory aid--a surgically implanted prosthetic device that bypasses the damaged inner hair cells and transduces an auditory signal into an electrical signal that stimulates the auditory nerve (Wilson, 2000). A cochlear implant provides patients who have profound hearing loss with useable forms of auditory stimulation. A typical multi-channel cochlear implant consists of a microphone that receives auditory input, a speech processor that uses one of several possible preset algorithms to VITEVITCH, PISONI, KIRK, HAY-MCCUTCHEON, AND YOUNT 192 process incoming auditory signals, and an array of electrodes that are surgically implanted into the cochlea to electrically stimulate the auditory nerve. Electrical stimulation of the auditory nerve by the implant results in the perception of spectral information via the tonotopic arrangement of the electrodes in the cochlea. The stimulation also provides durational and intensity information about the auditory signal (Wilson, 2000). The outcome measures of the effectiveness of cochlear implants in adults (across the several types of systems and several processing strategies) ranges from being able to follow a conversation on the telephone to being able to merely detect the presence or absence of sound (e.g., Blamey et al., 1987; Cohen, Waltzman, & Shapiro, 1989; Dowell, Mecklenburg, & Clark, 1986; Gantz et al., 1988; Skinner et al., 1991; Geier, Fisher, Barker, & Opie, 1999; Hollow et al., 1995; Holden, Skinner, & Holden, 1997; Staller et al., 1997). To examine whether cochlear implant users are still able to make use of phonotactic information to recognize spoken words, we used the nonwords of Vitevitch et al. (1997) with a slightly modified methodology. In the present experiment, participants were presented with bisyllabic nonwords varying in phonotactic probabilities and were asked to repeat the nonword as accurately as possible. After the repetition response, they heard the stimulus again but were asked to rate the goodness of each item as if it were a real word in English. Participants used a scale of 1 (“Bad sounding English word”) to 5 (“Good sounding English word”). If cochlear implant users are able to access phonotactic information, we would expect to find a difference in the ratings of the nonwords that is similar to that observed by Vitevitch et al. (1997). Specifically, nonwords with high-probability phonotactics should be rated as better sounding English words than nonwords with low-probability phonotactics by the cochlear implant users. Moreover, patients with better word recognition skills (as assessed by scores on the NU-6) may be able to more finely discriminate sound patterns and sequences varying in phonotactic probability and therefore would be more likely to use this more detailed information than those with poorer word recognition abilities (i.e., lower NU-6 scores). We further predicted that the ratings would reflect this difference in word recognition ability. Specifically, patients with poorer word recognition abilities should not be able to make fine-grained discriminations among segments and sequences of segments making it difficult to distinguish a real word from a nonword. This pattern would be expected in cochlear implant patients with poorer word recognition abilities. They would rate all of the nonwords as being "better words" than cochlear implant users with better word recognition ability or normal hearing listeners. For repetition accuracy, we predicted that if cochlear implant users were able to use phonotactic information, the accuracy with which the nonwords were repeated would also vary as a function of phonotactic probabilities. Specifically, nonwords with high phonotactic probability should be repeated more accurately than nonwords with low phonotactic probability, as in Vitevitch et al. (1997). Finally, we predicted that the cochlear implant users with better word recognition ability would repeat the nonwords more accurately than the cochlear implant users with poorer word recognition ability.