A Brief Taxonomy of Hesitation Phenomena

Hesitation phenomena are an integral part of speech, particularly unscripted, spontaneous speech. A crucial factor underlying all hesitation phenomena is the fact that all of them entail the delay of message transfer in some way. That is, if the phenomenon had not occurred, the communication of the speaker's message would have been faster. Yet, it is not easy to construct a clear categorization of the different types since there is characterization of hesitation phenomena in the literature has not been particularly consistent. Following is one categorization, but it should not be regarded as definitive or representing a consensus.

Silent pause
Unfilled pause
Silent pauses include intervals of silence within stretches of speech. However, not all silent intervals necessarily count as hesitation phenomena. Many of these silent pauses are simply juncture pauses (e.g., corresponding to, say, commas in writing), or pauses for articulatory reasons. Most researchers use some minimum cut-off point to distinguish between such naturally-occurring pauses and hesitation pauses. There is no clear consensus on the appropriate cut-off point and studies have used intervals as short as 200ms or as long as 800ms as a cut-off point. However, the shorter end of this scale (i.e., 200-300ms) is typical.
Filled pause
Filled hesitation
Vocalized hesitation
Filled pauses include instances in which speakers utter a syllable which typically consists of a centralized vowel as a nucleus and an optional nasal coda (e.g., in English, uh/um. Other phonemic forms of filled pauses are also recognized, but are typically phonemic combinations which do not coincide with well-recognized lexical items (e.g., in Japanese, e-to). In some cases, filled pause forms which correspond to demonstrative pronouns (e.g., Japanese ano-, Spanish este) are analyzed as filled pauses.
False start
A false start is when a speaker begins an utterance and then abandons it completely without finishing it.
A repeat is an immediate repetition of a sequence of one or more words. Repetition of the beginning of a word (i.e., stammering) is not usually included in this category as the cause is usually regarded as localized to articulatory problems rather than broader language production processes.
A restart is repetition of a sequence of one or more words at the beginning of an utterance. Typologically speaking, restarts could be considered a special type of repeat, yet many researchers distinguish the two.
A self-correction is when a speaker utters a sequence of one or more words that is to be understand as a replacement (i.e., correction) of the immediately preceding comparable sequence.
A lengthening is when a speaker elongates the articulation of one segment of a word. It often occurs on vowels, but can occur on consonants and even geminated consonants (by delaying the release).
Lexical filler
Lexical fillers include words or phrases that are conventionally used for the purpose of hesitation. The terminology is somewhat misleading, though, in that the name implies that other forms of hesitation phenomena are not lexical. There is, however, no consensus on this issue. Another point to made here is that some researchers do not include these as hesitation phenomena.

It is useful to note here that Levelt's model of speech production—one of the most influential models of speech production—sees all hesitation phenomena as occurring within the context of speech repairs. Some of these repair sequences occur overtly (after articulation; as in the case of self-corrections and false starts), while others occur covertly (before articulation; with silent and filled pauses, for example).