Using nonword repetition tasks as an experimental approach with both adults and children has become quite common in the past 10 to 15 years for studying lexical learning and phonological processing (e.g., Bailey & Hahn, 2001; Gathercole, Frankish, Pickering & Peaker, 1998; Munson, Edwards, & Beckman, 2005; Storkel, 2001; Vitevich & Luce, 2005). In her Keynote, Gathercole (2006) indicates that “The ability to repeat multisyllabic nonwords … probably represents the most effective predictor of language learning ability that is currently known” and that “nonword repetition … may also hold the key to understanding developmental disorders of language learning.” Her Keynote reviews many of the findings from a variety of nonword repetition studies with typically developing children and children with specific language impairment (SLI) or other language-related disorders. Despite the substantial benefits and interesting findings that nonword test paradigms have provided in addressing different issues, a number of questions and precautions should be kept in mind regarding such approaches. Researchers who routinely employ these procedures are likely to be well aware of these issues, and in fact, Gathercole points out some of the limitations. For those less familiar with nonword repetition tasks, it seems advisable to reinforce the need for a certain amount of caution in evaluating the results of such approaches.

The presumed benefit of using nonwords as stimuli in various types of experiments is that they involve limited linguistic processing (i.e., not to the same extent or possibly in the same manner that repeating real words does). However, it is often not clear specifically what differences are associated with nonword repetitions versus real words. Semantic processing is assumed to be largely lacking in nonword repetition, but certain stimuli are somewhat more similar to real words than others are in their phonotactic structure and/or other phonological characteristics, which is an important variable in many nonword studies. Although nonwords that are more wordlike may invoke some amount of lexical–semantic processing, it is difficult to quantify precisely the degree of wordlikeness of various stimuli (Bailey & Hahn, 2001), especially across individual subjects; and utilizing adult evaluations of wordlikeness when studying children is also a potential concern. A nonword such as “cloyper,” for example, could be judged to be somewhat similar to words such as “clover” and “clobber,” and perhaps “cloister” or other words for some subjects. Depending on the specific phonological characteristics of the nonwords and the lexical experience and sophistication of the subjects in a study (e.g., children with different lexicon sizes or typically developing vs. SLI children), varying and relatively unquantifiable amounts of lexical–semantic processing are likely to occur for different subjects and across different stimuli.

In addition, despite many well-designed experiments that have been conducted, it is not clear when utilizing nonword repetition procedures how accurately input, memory, and output factors can be distinguished in attempting to understand the findings of such studies. That is, in very general terms, nonword repetition tasks involve some amount of auditory, cognitive, phonological, and motoric processing, in that stimuli must be heard, remembered, and repeated. Given that speech output always constitutes the “end product” of this process, it is difficult to know with certainty if output errors that subjects make in an imitation task reflect whether they may have misheard the production of a stimulus to be repeated, it was heard correctly but then stored in short-term memory inaccurately, it was mispronounced for reasons unrelated to hearing or memory factors, or it was potentially due to some combination of these factors. For example, a child with a hearing loss (or possibly a child with normal hearing who was presented stimuli with a low signal to noise ratio) might produce output errors due to input problems, but also potentially because of memory and/or articulation limitations unrelated to his/her hearing status. Gathercole (2006) attempts to clarify some of these uncertainties, as have others, regarding the sequence of receiving input through producing output, but conclusive evidence indicating where in the imitation process malfunctions occur is difficult to obtain. One approach that seems to be somewhat helpful in this regard is to carefully examine the specific types of errors made at the output level utilizing various phonological analyses (e.g., Edwards, Beckman, & Munson, 2004), rather than only classifying a response as correct or incorrect, as has been done in some investigations. Particularly when studies involve relatively young children or children with SLI, evaluating their speech to determine what individual articulation errors they may manifest is important, so that such patterns can be factored into conclusions about their nonword repetition errors. That is, some output inaccuracies might not be a consequence of the test procedures, but may simply reflect a young child's individual speech patterns (e.g., Edwards & Lahey, 1998).

Another precaution concerning interpretation of data from nonword studies is that whatever adults or children do in the context of an experimental task may not reflect what they actually do in their normal lexical learning. For instance, the fact that a group of subjects shows evidence of statistical sensitivities/preferences for certain phonotactic patterns compared to others does not ensure that they typically utilize such propensities in their routine lexical–phonological processing. An analogous situation to consider is that in assessing speech production patterns and capabilities in children and adults over the past several decades, researchers have sometimes had subjects talk at faster than normal rates in different experiments. In fact, both children and adults have demonstrated that when required to do so, they can talk faster than they generally do (e.g., Smith, Sugarman, & Long, 1984). Nevertheless, under normal circumstances, they do not speak at those faster rates. Thus, there is a difference between the rates at which subjects typically speak compared to rates at which they are able to speak. Similarly, demonstrating that subjects show sensitivities to phonotactic patterns in nonword repetition tasks should be viewed cautiously, as they may not routinely utilize these abilities in the process of lexical or phonological learning. They may, of course; but in the absence of any conclusive evidence of this nature, it seems appropriate to be quite conservative when interpreting findings from such tasks.

One must also be careful about generalizing too broadly on the basis of group averages and statistical tendencies, especially in terms of how they might relate to the performance of individual subjects. Gathercole (2006) notes (as do others who utilize such procedures) that there is “substantial individual variation and developmental change,” but this awareness among investigators sometimes seems to be overlooked when conclusions are reached and models are developed. Some children may indeed have phonotactic sensitivities similar to those determined from group statistical results; however, others may not, or they may vary at different times in their development. In addition, given that some studies report statistically significant but nonetheless reasonably modest differences (e.g., 10–15%) between SLI and typically developing children or between high and low phonotactic probability, for example, one must be careful not to assume that such results are necessarily representative of each individual subject's performance. There is often considerable overlap in performance among subjects from different groups (e.g., Edwards & Lahey, 1998). To the extent a particular model assumes that certain phonotactic probabilities exist or that a particular processing sequence occurs, for instance, investigators must be careful not to conclude that all subjects possess the same abilities or demonstrate the same patterns. Thus, among a group of subjects, it is important to know how many of them show performance patterns consistent with group averages and for those who do not, how and to what extent they differ. When describing general findings or modeling a particular process, accounting for those subjects who do not reflect “average” or typical patterns is important.

In conclusion, research utilizing nonword repetition procedures has provided many new insights in recent years into the nature of word learning. It is likely that future investigations will continue to report additional intriguing findings that will further our understanding of this complicated process, particularly as data collection and analysis techniques continue to be refined. Further attempts to identify and clarify the nature of individual differences in response to such testing procedures will be one important direction to consider in greater depth, as will more fully understanding interactions among the specific perceptual, cognitive, and motor processes involved in such tasks.