The study of development of vocalization behavior has long been dominated by two trends. In the first trend, interaction between infants or children and parents has been observed, usually by developmental psychologists. Categories of coding have been few, generally 2-5, in these interaction studies, and the paucity of categories has limited the interpretive power of the work. On the other hand, interaction studies have tended to provide sophisticated monitoring of observation reliability for the small number of categories employed in the work.

In the second type of research, infant or child vocalizations have been observed independent of interaction, in studies often performed by linguists or psycholinguists. The number of categories of coding has tended to be large. For some such coding schemes, each event can potentially be coded in terms of hundreds of possibilities. The complexity of linguistically-sophisticated coding systems inevitably complicates reliability assessment. In order to evaluate reliability quantitatively, much more effort is needed than in studies with few categories, and as a result there has been a lack of appropriate attention to reliability in linguistically oriented studies of vocalization.

Recently, the problem of reliability assessment has been intensified in linguistically-oriented studies of vocalization because schemes of coding for vocalization have been complicated even further in response to the recognition that patterns of infant vocal activity are organized hierarchically in syllables, utterances, phrases, and even higher order rhythmic units [1,2,5]. The prosodic characteristics of these higher order units is of increasing interest because it appears to reveal that infants even in the first months of life possess the core features of the rhythmic organization of prototypical mature speech. The infant capability is manifest, for example, in similar average durations of hierarchical units for infant vocal sequences and for clearly produced mature conversational speech, or for such prototypical rhythmic forms as nursery rhymes.

Computer-based observational coding and training offers new possibilities for the establishment of new coding schemes, the enhancement of reliability and reliability assessment and for development of more compatible cross-laboratory procedures. The authors of the paper have been working for years on the development of such improved procedures. In particular, the paper will report on efforts in which we are developing a library of digitized infant vocalizations, with coding recommendations to be utilized in interactive computer-based training. The examples focus on categorization at three levels in the rhythmic hierarchy of vocalization (syllable, utterance, phrase) as well as on many features of individual syllables and their presumed primitive segmental (consonant/vowel) substructure. The examples have been coded by two 'gold standard' observers, experienced coders of infant vocalizations and contributors to the literature on coding. For each utterance or utterance sequence in the database, reference codings are specified in terms of an infraphonological model [3]. The codings are explicated in accompanying documentation. In many instances binary categories of coding pertain to continuous variables, and consequently some utterances may represent clear examples of a categorical type (pertaining to an end of the continuum to which the type belongs, far from the boundary between types), while others may represent relatively ambiguous cases (pertaining to the middle of the continuum, near the boundary between types). Both clear and relatively ambiguous examples of various categorical types are available, clearly marked as such in the database, along with explication.

The training procedures are designed to enhance reliability and insure possible consistency in criterion setting for categorization. The method is based on presentation of examples in a systematic format within LIPP (Logical International Phonetics Programs) files [4], with automatic access to the database and the codings of the gold standard observers. After introductory training with each dimension of coding, learners are subjected to tests of their coding also conducted within LIPP. In this way it is possible to insure that the basic categories of coding are treated in the same way by observers in various laboratories. The LIPP system, with its specialized analysis language, offers automatic comparisons of coding data from various observers and so provides a basis for systematic tests of reliability. Changes in coding conventions can also be implemented conveniently within this software framework. The work is pursued in the attempt to integrate the sophistication of reliability assessment from studies in developmental psychology with the sophistication of linguistically-oriented vocalization coding schemes.

References