The study of disfluencies belongs under the psycholinguistic approaches to researching interpreting. Disfluencies can occur at any level of speech planning and execution (Levelt, 1989), they can either signal errors related to conceptual or grammatical planning, or they can be connected to overt or covert monitoring (Lickley, 2015). A possible obstacle to research and comparing research results is that there is neither an accepted standard definition, nor a universal taxonomy of disfluencies.

An early definition states that a disfluency is “an involuntary deviation from the speaker’s current phonological, grammatical, or lexical intention” (Boomer & Laver, 1973, p. 123). A more recent definition, by Gyarmathy, states that disfluencies are “those phenomena of speech that in a sense interrupt the articulatory and perceptual fluency of spontaneous speech, and signal the uncertainty of the speaker in speech planning, or refer to the problems or malfunctions in the speech planning process, and no pragmatic function can be associated to them”1 (Gyarmathy, 2015, p. 13). Gyarmathy classifies disfluencies into two broad categories: error-type disfluencies (ETDs) and disfluencies rooted in the uncertainty of the speaker. The category of ETDs includes Freudian slips, malapropisms, grammar errors, blends, false word, tip of the tongue errors (TOTs), ordering problems (perseveration, anticipation, and metathesis) and simple slips (addition, deletion, exchange). The second category of disfluencies, which are rooted in the uncertainty of the speaker, includes pauses, hesitations, pauses within the word, repetitions, fillers, lengthenings, and restarts. The most frequent typology of disfluencies (Roberts et al., 2009; Lickley, 2015) is summarized in Table 10.1.

The most widely accepted model of speech production is that of Willem Levelt (1989). The strength of Levelt’s model is that it describes the complete process of speech production and articulation, from the intention of speaking to articulation. The model provides a synthesis of models and theories of speech production, however, it mostly draws on speech production experiments related to English and Dutch and does not always take into consideration differences between languages.

Levelt’s model of speech production, “A Blueprint for the Speaker” (Levelt, 1989, p. 8–22) is made up of processing components and knowledge stores. The first processing component is the Conceptualizer, this is where pragmatic and semantic planning happens, the result of these processes is the preverbal message. The second processing component is the Formulator, where grammatical and phonological encoding take place. The resulting phonetic plan is fed into the third component, the Articulator, which is responsible for the motor execution of speech. The knowledge stores include the Lexicon and the knowledge store containing the discourse model, situation knowledge, and the encyclopedia.

In Levelt’s model, speech production has two stages: macroplanning and microplanning. During macroplanning, the Speaker decides on what information he or she would like to express to the Listener, in other words, the Speaker’s illocutionary intentions are formed (Levelt, 1989, p. 123). During the second stage of macroplanning, the Speaker arranges the information to be expressed in some order (Levelt, 1989, p. 138). Macroplanning is followed by microplanning, in the course of which the conceptual structure is translated into a linguistic structure. Based on the preverbal message, two concurrent processes take place during grammatical encoding. The Formulator activates lemmas belonging to the concept of the preverbal message from the Lexicon. In addition, the surface structure of the utterance is formed based on the syntactic features of lemmas. Levelt opines that the surface structure resulting from grammatical encoding is lexically determined, in other words, the meaning of the message generates the surface structure through lemmas. This means that the syntactical and morphological characteristics of activated lexical elements lead to the surface structure. The process of grammatical encoding manifests itself and can be researched through grammar errors.

During the process of phonological encoding, a phonetic or articulatory plan is retrieved or built for each lemma and for the utterance as a whole. This involves three distinct processing levels. The first is the morphological level, its input is the lexical form, which is the Lexicon’s information of an item’s internal composition, including lemma information, together with information on its morphology and phonology. The output of the first stage of phonological planning is the morphological and metrical composition of a word. The second level is the phonological assembly of a word, its input is morpheme and metrical information, and its output is the composition of the word’s segments. The third level is phonetic planning, when segments are forwarded to the phonological encoder, which produces a phonetic plan for syllables.

All of this information is fed to the Articulator, where the phonetic plan is executed by the muscle movements of the respiratory, the laryngeal and the supralaryngeal systems. Before articulation, the phonetic plan is temporarily stored in the Articulatory Buffer, in order to balance the differences of speed of planning and articulation.

Levelt’s model also includes self-monitoring (Levelt, 1989, p. 459–499). When a Speaker detects an error related to the meaning or form of his or her internal or overt speech, the Speaker may stop and start a repair. The Speaker will not wait for finishing a syntactic, lexical or phonetic unit. In the moment after stopping or halting, an editing expression can be detected in speech. Self-monitoring and self-repairs largely depend on the circumstances of speech production. During semantic planning and grammatical encoding, corrections are possible within the Conceptualizer and the Formulator, while monitoring of phonological-phonetic encoding happens through the speech perception system (Csépe et al., 2008, p. 106).

In his revised model (Levelt, 1999), Levelt differentiates between the semantic/syntactic system and the phonological/phonetic system and describes speech production as the result of the cooperation of these two systems. The semantic/syntactic system carries out conceptual planning and grammatical encoding, while the phonological/phonetic system carries out morphological and phonological encoding, phonetic encoding and articulation. These two systems are complemented by three knowledge stores, namely knowledge of the world, the mental lexicon, and the syllabary. World knowledge plays a role in conceptual planning, while the mental lexicon is connected both to the semantic/syntactic system and the phonological/phonetic system. Lemmas are needed in grammatical encoding while word from information is used in morphological and phonetic encoding. The syllabary is a store of automatic articulatory movements that are needed to form syllables, and it helps in phonetic encoding (Levelt, 1999).

Disfluencies signal malfunctions at different levels of speech production. Through their analysis, one can refer back to the origin of a problem that manifests itself as a disfluency on the surface structure. The following summary is based on Gyarmathy (2015). Freudian slips and malapropisms signal malfunctions of conceptual planning. Grammar errors are evidence of problems related to grammatical encoding during speech production. Blends result from the mixing of two competing, activated forms in the mental lexicon. False word errors and tip of the tongue errors signal malfunctions during lexical access. Ordering errors are related to the problems in the control of timing of the planning and execution of speech.

Slips are errors related to articulation. The origin of false start errors is not possible to pinpoint, they can be connected to any level of the speech production process.