With great clarity Ann Peters has summarized the way we have been viewing ‘fillers’ from the time they were considered a nuisance for seriously studying the development of language; the different roles they fulfill, depending on the developmental stage; and important open questions that should bring our understanding of fillers ahead. She notes that the reason for having ignored fillers for quite a while is that ‘they do not fit neatly into linguist's notions about ‘modules’ of language because at the very least they straddle preconceived boundaries.’ In fact, it is not a coincidence that researchers began to develop an interest for such neglected entities, as the interfaces of grammar began to play a central role in our understanding of language architecture. Being at the crossroads of phonology, morphology and syntax, fillers resist a rigidly compartmentalized view of language.

Although interest in fillers has grown amazingly, we definitely need more detailed descriptions of filler production and development, in order to be able to distill individual and language specific trends. On this note, I want to emphasize some of the general features of early fillers and argue that they neither support nor justify the assumption of a pregrammatical stage. On the contrary, their very presence argues against it.

Fillers appear during the early grammaticalization process, roughly between the one- or two-word phase and the earliest grammar, in a variety of languages. Peters accurately points out there have been theoretical and methodological difficulties when trying to integrate the filler phenomenon in language acquisition theory, probably because fillers do not fit neatly into linguistic categories, nor a structure of ‘modules’. Her note is a major contribution to the solution of those problems. She gives us a rich review that can guide any reader – specialised or naive – to deep, critical insights and to the relevant research literature. In addition to this valuable contribution, she points the way to future work. The very richness makes it difficult to add any interesting comments. So these remarks will be limited to complementary concepts and data.

Peters sets up a very useful initial structure for classifying elements which are not yet well-defined morphemes, especially those that seem to be modelled phonologically on adult functors. Some of these primitive elements (pre-morphological items) have neither semantic content nor distributional properties that correspond to any adult morpheme, while the more advanced (protomorphological) items have begun to take on some of the syntactic/semantic characteristics of particular classes of adult-language morphemes.

Ann Peters' interesting paper raises a number of questions in my mind. One of these arises from ‘the assumption that at any given point in development, children either do or do not ‘have’ the functional categories that underlie syntax.’ This dichotomy may be too simplistic. A number of researchers have argued over the past few years that during the Optional Infinitives/OI stage of Wexler (1994) children may ‘have’ UNDERSPECIFIED functional categories. For example, Schütze & Wexler (1996) and Schütze (1997) argue that the category INFL at the OI stage may be underspecified in respect of its tense/agreement features: so, for example, in He cries, INFL is specified for both tense and agreement (with agreement triggering nominative casemarking of the subject he); in He cry (e.g. in reply to a question like ‘What did the baby do?’) INFL is specified for agreement but not tense; and in Him cried, INFL is specified for tense but not agreement. Given Chomsky's (1995) view that categories are sets of features, it follows that an underspecified child INFL constituent cannot in principle represent THE SAME CONSTITUENT as an adult INFL fully specified for tense and agreement (and perhaps mood/finiteness as well), but rather is a proto-INFL. This in turn suggests that children have to learn to build up feature complexes associated with functional categories ‘one feature at a time’ – and that the OI stage represents a period when certain features of INFL are taken to be optional. In other words, it may be that Peters' observation that ‘children must construct their grammatical categories on the basis of gradual learning’ is as true of functional categories as it is of substantive categories. On this view, functional categories MATURE. Indeed, this conclusion seems to be a matter of virtual conceptual necessity if we follow Chomsky (1999) in positing that learners have to ASSEMBLE lexical items from sets of features provided by the Language Faculty, and if we assume that feature-assembly is not instantaneous.

In her note Ann Peters makes a remarkable effort in bringing together data on fillers reported to occur in different languages. This is a difficult enterprise since these data have been obtained with different methodological tools and are anchored in different theoretical approaches. Peters tries to organize them under a loose but common underlying theoretical framework in which phonoprosodic, lexical, syntactic and morphological aspects of acquisition are considered as inherently interrelated and intertwined. Within this framework she is able to consider as related fillers that seem to stand for different kinds of language units (free and bound morphemes, but also amalgams and unidentifiable chunks that seem to stand for content words). She points out that fillers may be at the crossroad of these different kinds of traditional layers of language. As such, they may constitute traces of different kinds of work from the child's side and may have a role in changing the status of the child's knowledge in these domains.

Peters' Note highlights the tight interconnection between phonological advances and the child's discovery of morphosyntactic structure. Taking a constructivist view, or assuming no advance knowledge of what linguistic principles or parameters may be expected, investigators like Peters & Menn (1993) or Veneziano & Sinclair (in press) come to the conclusion that children must begin with the surface regularities of language, only gradually familiarizing themselves deeply enough, through use, with the make-up (as well as the situational cooccurrences) of commonly heard words and larger lexical patterns or constructions to begin to develop multi-level linguistic knowledge, i.e., to discover morphosyntax. In commenting on Peters' stimulating note on filler syllables we would like to draw on our cross-linguistic data for the late single word period to attempt to derive some further hints as to the difference input prosody may make and the kinds of transitional phenomena one may find.

All the comments were thoughtful, interesting, supportive of further investigation into this small but complex arena of language acquisition, and suggested new ways to extend the enterprise. Not having space to discuss all the interesting issues to the length they deserve, I will synthesise the highlights of an overall picture of both agreement and questioning.