Fox, Barbara, Maschler, Yael, and Uhmann, Susanne. (2010). A cross-linguistic study of self-repair:...

A cross-linguistic study of self-repair: Evidence from English, German,and Hebrew

Barbara A. Fox a,1,*, Yael Maschler b,1, Susanne Uhmann c,1

aDepartment of Linguistics, 295 UCB, University of Colorado, Boulder, Boulder, CO 80309, United StatesbDepartment of Communication, University of Haifa, Mount Carmel, 31905 Haifa, IsraelcUniversity of Wuppertal, Geistes- und Kulturwissenschaften, 42119 Wuppertal, Germany

1. Introduction

This paper explores same-turn self-repair in three languages: English, German and Hebrew. Same-turn self-repair is theprocess by which speakers of a language stop, abort, repeat, or alter their turn before it comes to completion.

For some time it has been known that same-turn self-repair (hereafter: self-repair) is highly organized. That is, self-repairis not produced randomly but is highly patterned, both phonetically and morpho-syntactically (Jesperson, 1924; see alsoMaclay and Osgood, 1959; Hockett, 1967; Schegloff et al., 1977; Schegloff, 1979; Levelt, 1982). What has not been knownuntil recently, however, is that the organization of self-repair varies from language to language. For example, whilerepetition of an entire clause occurs in English with some frequency, in Japanese it is extremely rare (Hayashi, 1994; Foxet al., 1996). Moreover, while replacement of one bound morpheme with another occurs in Japanese, Korean and Finnish, ithas not to date been found in English (Hayashi, 1994; Fox et al., 1996; Yang, 2003; Karkkainen et al., 2007).

Journal of Pragmatics 42 (2010) 2487–2505

A R T I C L E I N F O

Article history:Received 9 February 2008Received in revised form 2 November 2009Accepted 10 February 2010

Keywords:Self-repairTypologyDiscourse-functional syntaxComparative syntax

A B S T R A C T

This paper presents the results of a quantitative analysis of recycle and replacement self-repairs in English, Hebrew and German. The analysis revealed patterns of similarities anddifferences across the languages. Beginning with patterns of difference, we found first thatEnglish and Hebrew speakers engage in simple recycling about two-thirds of the time,while German speakers make less frequent use of simple recycling. Second, we found thatEnglish speakers frequently recycle back to the subject pronoun of a clause, while Hebrewand German speakers makemuch less use of subject pronoun as a destination of recycling.Third, we found that Hebrew and German speakers recycle back to prepositions muchmore frequently than do English speakers. With regard to similarities across the threelanguages, we noted that all three languages used function words as destinations ofrecycling more often than content words, while replacing content words at adisproportionately high rate. We claimed that entrenched word order patterns play acrucial role in explaining the facts we have observed; patterns of morphologicaldependence across collocates also shape self-repair practices in these languages. Thisstudy is thus further evidence of the shaping role that morpho-syntactic resources have onthe self-repair practices of a speech community.

! 2010 Elsevier B.V. All rights reserved.

* Corresponding author. Tel.: +1 303 492 6305; fax: +1 303 492 4416.E-mail addresses: [email protected], [email protected] (B.A. Fox),[email protected] (Y.Maschler), [email protected]

(S. Uhmann).1 These authors contributed equally to this research.

Contents lists available at ScienceDirect

Journal of Pragmatics

journa l homepage: www.e lsev ier .com/ locate /pragma

0378-2166/$ – see front matter ! 2010 Elsevier B.V. All rights reserved.doi:10.1016/j.pragma.2010.02.006

mailto:[email protected]




https://ssl.haifa.ac.il/science/journal/,DanaInfo=www.sciencedirect.com+03782166

https://ssl.haifa.ac.il/10.1016/,DanaInfo=dx.doi.org+j.pragma.2010.02.006

Cross-linguistic variation in patterns of self-repair are particularly significant because previous studies suggest arelationship between the typological characteristics of individual languages and patterns of self-repair (Fox et al., 1996;Fincke, 1999; Uhmann, 2001; Wouk, 2005; Karkkainen et al., 2007). It appears that a range of typological features, such asword order, favored anaphoric devices, morphological complexity of words, degree of syntactic integration, and presence orabsence of articles and adpositions, influence self-repair in a variety of ways.

The current paper presents results from a comparison of self-repair in English, German and Hebrew. Our results suggestthat in spite of the close genetic relationship between English and German, typological characteristics that distinguish themproduce markedly different self-repair patterns. Our findings lend support to the claim that it is typological features, ratherthan, for example, genetic closeness, which produce patterns of self-repair.

In this paperwewill focus on self-repair which contain simple recyclings or simple replacements. Some English examplesof the kinds of data included in our study are given below:

(1) Hey would you like a Trenton::, (.) a Trenton telephone directory

(2) and the the moo- thing was the Dark at the Top of the Stairs

Example (1) illustrates simple recycling, that is repetition of words already produced by the speaker without any otherprocess involved. Example (2) illustrates simple replacement of a word; here the speaker replaces what is likely thebeginning of the word movie with the word thing.

Our larger databases include instances of more ‘elaborate’ repairs, with pre- and post-framing, additions and deletions,and complex combinations of repair types; by limiting the study to two repair types, we hope to position ourselves mosteffectively to understand our findings.

The current paper focuses on the syntactic constituents which are involved in recycling and replacement repairs in ourthree languages, and the relationships between those syntactic constituents and the morpho-syntactic organization of eachlanguage. The goal of the paper is to present these findings, to offer explanations for them, and to suggest implications of thedifferences for the organization of self-repair and for the organization of morpho-syntax more generally.

Although there are interesting differences in the frequency of each repair type across the languages, due to limitations ofspace a discussion of those differences will not be offered here.

The paper is organized as follows. Section 2 provides a description of the data and methods of the study; section 3presents our findings. Section 4 presents discussion, and section 5 concludes the study.

2. Data and coding

2.1. Data collection

The data for this study come from three corpora, one for each language – English, German, and Hebrew. Each corpusconsists of several audio-taped (and, for English, video-taped) casual face-to-face conversations among friends and familymembers, in interactions among 2–5 participants per interaction.

Self-repair tokens were collected as part of a larger project on self-repair (Fox and Wouk, 2003) according to theguidelines for that project: up to 100 instances of self-repair were taken from each interaction. The total numbers ofinstances for English was 500, for Hebrew it was 250, and for German it was 231. For the English corpus, thedata represented the speech of 19 speakers, across 6 interactions, totaling approximately 3 h of interaction. For theHebrew corpus, this procedure resulted in investigating 64 min of discourse among a total of 51 speakers distributedacross 23 different interactions from the Haifa Corpus of Spoken Israeli Hebrew (Maschler, 2009). For the Germancorpus, this procedure resulted in investigating 130 min of discourse among a total of 7 speakers across 6 differentinteractions.

For the purposes of the current report, we have excluded all instances of self-repair that occurred in the environment ofoverlapping talk from another participant. We made this decision in order to achieve greater simplicity in the discussion offunctions of self-repair, especially recycling.

2.2. Data coding

2.2.1. Coding according to repair typeAll self-repair tokens were classified according to repair type: simple recyclings and simple replacements. An asterisk

denotes the point of repair initiation, boldface denotes the item replaced and the item replacing it, and boldface + italicsdenotes the items recycled.

An instance of repair was treated as a replacement if a word (or multiple words) was replaced. Consider example (3)below. In this utterance the speaker produced thewordwriting, paused for half a second following the articulation ofwriting,replaced it with spray painting, and then proceeded to complete the question:

(3) What was this I heard about them going up to Monarch and writing* (0.5) spray painting something on Monarch?

B.A. Fox et al. / Journal of Pragmatics 42 (2010) 2487–25052488

We also included in replacements instances in which a speaker mispronounced a word and then re-produced it with amore appropriate pronunciation, as in example (4) below. In this example, the speaker produces the beginning of the name ofa sports team, but produces a pronunciation that does not fit the name of the team (Jo-). He then replaces that pronunciationwith the more appropriate pronunciation (Jets):

(4) is there a more perfect guy to fit the Jo-* Jets organization?

Instances in which the speaker repeats a word base and replaces affixes on that base were also included as simplereplacements. The change from the Hebrew verb haya (‘was’, masc.) to the verb hayta (‘was’, fem.), with recycling of the base,in utterance (5) below, for instance, was coded as Replace2:

An instance of repair was treated as a simple recycling if the speaker repeated one or more words. Consider example (6).In this example the speaker produces you can get a, pauses for 0.7 s, and repeats you can get a:

(6) I mean it- th’t’s pretty small you can get a* (0.7) you can get a* ah:: (1.1) you know you can get a pretty smallone [now.

Instances involving the addition or deletion of an element and then recycling of a word already produced were excludedfrom the study. Thus utterances like (7) were not included in the study:

(7) the interesting thing about the third one (0.4) is that it was made in* I think it was made in 1990.

Self-repair tokens which were not simple recycling or simple replacements were coded as ‘Other’, and will not beexplored further in this study. Cases in which a structure was completely aborted and a new one begun (type G in Fox andJasperson, 1995), and placeholder examples (Fox et al., 1996:206) were ignored. Simple recycling produced in overlap werealso ignored.

2.2.2. Coding according to syntactic categoryWe coded our data for two features3: the syntactic category of the replaced item(s) in replacements; and the syntactic

category of the first item to be repeated in recycling repairs. We refer to the latter as ‘destination of recycling’.Thus, e.g., in example (3)

(3) What was this I heard about them going up to Monarch and writing*(0.5) spray painting something on Monarch?

2 Transcription Method: The Hebrew data uses the following transcription notations: Each line denotes an intonation unit (Chafe, 1994) and is followedby an English gloss. In the cases in which this gloss is not close enough to an English utterance, it is followed by a third line supplying a usually literal (butsometimes functional) translation. Utterances under consideration are given in boldface. Transcription basically follows Chafe (1994), with a few additions.Conventions are as follows:

. . . – half second pause (each extra dot = another 1/2 s)

.. – perceptible pause of less than half a second(3.22) – measured pause of 3.22 s, – comma at end of line – clause final intonation (‘more to come’). – period at end of line – sentence final falling intonation? – question mark at end of line – sentence final rising intonation! – exclamation mark at end of line – sentence final exclamatory intonationø – lack of punctuation at end of line – a fragmentary intonation unit, one which never reached completion.- one hyphen – cutoff at repair initiation– – two hyphens – elongation of preceding vowel sound[square bracket to the left of two consecutive lines indicates[overlapping speech, two speakers talking at oncealignment such that the right of the top line

is placed over the left of the bottom line indicates latching, no interturn pause/??????/ – transcription impossible/words within slashes/ indicate uncertain transcription[xxxxx] – material within square brackets in the gloss indicates exuberances of translation (what is not there in the original).{in curly brackets} – transcriber’s comments concerning paralinguistics and prosody,which do not have an agreed upon symbol in this transcription system.

3 For the current studywe did not code specifically for syntactic category of the item inwhich repair was initiated. For example, in example (6) we did notcode for the fact that repair is initiated immediately after the indefinite article. For this study we focused on the syntactic category of the item(s) replaced,and on syntactic category of the item that was the destination of recycling.

B.A. Fox et al. / Journal of Pragmatics 42 (2010) 2487–2505 2489

we coded the item replaced as a verb. In example (6)

(6) I mean it -th’t’s pretty small you can get a* (0.7) you can get a ah:: (1.1) you know you can get a pretty smallone now

the destination of recycling was Subject Pronoun (you). It is important to remember that our discussions of recycling alwaysconcern the destination of recycling rather than theword inwhich recycling repair was initiated (although those are inmanycases one and the same word).

3. Recycling and replacement

The central finding of the study is an association between recycling and closed-class function words on the one hand andreplacement and open-class content words on the other. That is, we have found that in all three languages speakers tend torecycle back to function words much more frequently than they recycle back to content words; and we have found that in allthree languages content words are over-represented in replacement repairs (based on their total frequency in the repair data).Table 1 above gives the data for recycling and replacement repairs for function words and content words in each language.

We can see from Table 1 that recycling is associated with function words in all three languages, and that content wordsare over-represented in replacement repairs in all three languages. In English, only 15% of the destinations of recycling arecontent words, while 36% of replaced items are content words. In Hebrew, 16% of the destinations of recycling are contentwords, while 48% of replaced items are content words. And in German, 22% of destinations of recycling are content words,while 50% of replaced items are content words. Thus, for all three languages, the percentage of recyclings that begin withcontentwords is not higher than 22%, while the percentage of replacements that replace contentwords is between one-thirdand a half, with English having the lowest percentage and German the highest. These findings corroborate earlier studies onEnglish and German, according to which function words are repeated far more often than are content words (Maclay andOsgood, 1959; Lickley, 1994; Rieger, 2003), but this is the first study documenting this quantitatively and for Hebrew aswell,and relating also to replacement.

Fig. 1 below is a visual representation of these findings. Fig. 1 gives the percentage, for each language, that speakersrecycle back to a functionword or a contentword (fw = functionword; cw = contentword). It is clear that all three languagesrecycle function words much more frequently than they do content words.

Fig. 2 provides similar data for replaced items. As can be seen from Fig. 2, function words are less frequent as replaceditems than they were as recycled items, and content words are muchmore strongly represented as replaced items than theywere as recycled items. This pattern holds for all 3 languages.

Table 1Distribution of repair type by syntactic class for each language.

English Destination of recycling Replaced item Total

Function 95 (85%) 23 (64%) 118Content 16 (15%) 13 (36%) 29

Total 111 36 147

Hebrew


Total 128 27 155

German


Total 98 44 142

Fig. 2. Replaced items and syntactic class for all 3 languages.

Fig. 1. Destinations of recycle repairs and syntactic class for all 3 languages.


The first goal of the paper is to explain these correlations. The second goal of the paper is to describe and explain theparticular patterns of syntactic constituencies involved in recycling and replacement repairs across the three languages.

We turn now to an examination of recycling repairs in our three languages.

3.1. Recycling repairs

Aswe have just seen, all three languages exhibit a tendency to recycle back to functionwords. In English, 85% of recyclingrepairs recycle back to a function word; for Hebrew it is 84%, and for German it is 78%.

3.1.1. Function words in recyclingIn Table 1 we saw that in all three languages content words are the destination of recycling roughly 15–22% of the time,

compared to78–85%of the time for functionwords. This fascinatingdistribution canbe explainedby the functions of recycling.It has been claimed for English that recycling is one resource for delaying next item due (cf. Schegloff, 1979; Fox et al.,

1996). In English, function words tend to precede content words, and thus function words can serve as devices on which toproduce prospective repair if there is troublewith an upcoming contentword (cf. Fox et al., 1996), in that recycling delays theproduction of that content word. For example, subjects are overwhelmingly pronominal in English (function words), andsubject pronouns tend to precede verbs, which are content words. So recycling a subject pronoun—a very common practicein English conversation, as we’ll see below—can be a device formanaging trouble with an upcoming verb. Similar argumentscan be made for determiners, and prepositions—all function words that typically precede content words. Recycling afunction word that precedes a content word is one way of delaying the production of that content word.

The same argument clearly holds for German. Determiners precede nouns, prepositions precede their noun phrases,auxiliaries precede their verbs, connectives come at the beginnings of clauses, and subject pronouns sometimes occur at thebeginnings of clauses. So recycling of these function words is an obvious device for delaying the content word(s) which theyproject.

Hebrew also has function words that project an upcoming content word, though perhaps not to the extent exhibited byEnglish and German. There are prepositions, and a definite determiner (though not an indefinite one), both of which precedetheir nouns. However, if a definite noun ismodified by an adjective, the adjective follows the noun andmust also be precededby a definite determiner. Thus, not all definite determiners precede their nouns. Demonstrative determiners, too, follow theirnouns. Conjunctions do occur at the beginnings of clauses, discourse markers occur at the beginnings of conversationalactions (Ford and Thompson, 1996) they connect, and subject pronouns can begin a clause, although not nearly as frequentlyas in English. Auxiliaries are not employed as frequently in Hebrew, but when they are, they precede their main verbs.

The discussion above accounts for the many instances in our three languages in which a speaker produces a functionword, recycles it, and then produces the sought-for content word. An example from English illustrates this pattern:

(8) and the* the moo- thing was the Dark at the Top of the Stairs

In this example the speaker produces a definite determiner, recycles it, and produces a first attempt at the sought-fornoun. (Interestingly, she then goes on to replace the selected noun with a much less specific one).

However, it is also possible for the speaker to initiate repair in a lexical item and then recycle back to another word. Forexample, a speakermay initiate repair in a contentwordand then recycleback toa functionword, as in example (1) fromabove,in which the speaker has produced the first component of a compound noun and recycles back to the indefinite article:

(1) Hey would you like a Trenton::,* (.) a Trenton telephone directory

Why do speakers recycle back to a function word in examples like this (note that they do not always—we have instancesin each language in which the content word is recycled on its own)? We propose that speakers recycle back to a functionword—often the one that started the local phrase, but in some cases further back—to give extra beats of delay for continuingthe turn. In fact, at least in the English data, recyclings that are initiated in content words and that do not recycle back to afunction word sometimes exhibit quite long silences before the recycling, as in (9) below:

(9)

Jenn: Ne[ver having

Bett: [(What.)

Jenn: a b:* (1.0) break.

In this example, the speaker closes her lips for the (audible) production of [b], and she holds that closure for over onesecond before releasing it into the rest of the word. Languages thus provide a range of resources for delaying next item (orsound) due; recycling back to the beginning of a phrase or clause may be one of those resources, while silence or a filler maybe others. And in fact, it appears that some languages prefer one device over others—for example, aswill become clear below,English has a strong preference for recycling more than one word, while German and Hebrew tend to recycle single words(and there are reports that German speakers may prefer fillers over recycling; see Rieger, 2003). And in section 5 we willpresent evidence from a language inwhichmodifiers do not precede heads, that the associationwe have found here between


recycling and function words does not apply; in other words, this association is specific to a set of typologically similarlanguages. Patterns of repair are thus shown to be related to morpho-syntactic typology.

It is also quite possible that different repair devices are associated with different action-types. To date little research hasbeen done to explore this possibility, and we will not pursue it further here.

3.1.1.1. A closer look. If we now look in detail at syntactic class and recycling for each of the three languages, wewill see theseassociations between recycling and function words borne out. Table 2 presents the relevant data for English.

We begin with a discussion of the English data. Subject pronouns in English make up 43% of all destinations of recycling.No other syntactic category comes close to that level. Determiners make up 12% and prepositions make up 9% of alldestinations of recycling. Nouns, verbs and adjectives are quite infrequent, each constituting 4% of all destinations ofrecycling. These patterns make sense given what we have said about the functions of recycling.

Clearly, themost striking pattern in the English data concerns the extremely high rate of subject pronoun as a destinationof recycling. Examples (10) and (11) illustrate the practice in English of recycling back to a subject pronoun:

(10) but it was-* (.) it was bad

(11) You’re li-* you’re like o- o- operating in terms of a m- of a slightly more organized life, than you might

We have noted elsewhere this strong tendency that English speakers have to recycle back to the subject pronoun (Fox andJasperson, 1995; Fox et al., 1996), but this is the first quantitative evidence for the claim.Moreover, as has been pointed out inprior studies (Fox et al., 1996; Wouk, 2005; Fincke, 1999), and supported by the findings presented in the current study,English is quite unique in its great enthusiasm for this particular practice.

As has been suggested elsewhere (Fox et al., 1996;Wouk, 2005; Karkkainen et al., 2007), English speakers’ predilection forrecycling back to subject pronouns arises fromanunusual set of factors. First, English speakers produce overt subject pronounsin nearly every clause in conversation (with a small class of exceptions; see Oh, 2005), in striking contrast to most otherlanguages whose conversational patterns have been studied. As we might say, you can’t recycle what you haven’t got, sospeakers of most languages don’t recycle back to subject pronouns simply because subject pronouns are not commonlyproduced in those languages. Second, subject pronouns typically occur in English at the beginnings of utterances, in contrastwith other languages (likeGerman)where subject pronounsmay take up a range of other locations in the utterance). Third, thehigh level ofbondingbetween the subjectandtheverb inEnglish—manifestedat least inpart by thecliticizationofauxiliaryandcopula forms to subject pronouns—makes the subject-verb complex a deeply entrenched grammaticizedunit (cf. Bybee, 2006),one that is rapidly available to the speaker (cf. Rieger, 2003). Some scholars have even suggested thatmost subject pronouns inEnglishare themselves clitics, so thatmost subjects forma tightunitwith their verb (seeDixon,2007). Fourth, thebeginningof aturn is a moment of heightened interactional significance especially with regard to turn-taking in English (Sacks et al., 1974;Schegloff, 1987), and there are a variety of reasons for re-doing the beginnings of turns (for example if the first attempt wasproduced in overlap; see Schegloff, 1987; or to achieve recipient gaze; see Goodwin, 1979, 1981). And fifth, it has been arguedthat speakers tend to recyclemore at thebeginnings of complex syntacticunits (Clark andWasow,1998), and subject pronounsin English occur at the beginnings of clauses/sentences,which are very complex syntactic units. There is no other language thatwe are aware of—including the sister Germanic language, German—that displays all of these characteristics, making theprevalence of subjectpronoun recyclingquiteunique toEnglish.Wewill seebelowthatGermanspeakers donot recycle back tosubject pronouns with the same frequency as English speakers do (cf. Rieger, 2003).

Table 2Syntactic category and repair type for English.

English Destination of recycling Replaced item Total

Function wordsSubject pronoun 48 (43%) 6 (17%) 54Determiner 13 (12%) 4 (11%) 17Wh-word 10 (9%) 0 (0%) 10Preposition 10 (9%) 3 (8%) 13Aux 5 (4%) 0 (0%) 5Existential 2 (2%) 0 (0%) 2Copula 2 (2%) 3 (8%) 5Connective 2 (2%) 1 (3%) 3Other 4 5 9

Content wordsAdverb 2 (2%) 2 (5%) 4Verb 5 (4%) 4 (11%) 9Adjective 4 (4%) 2 (5%) 6Noun 5 (4%) 7 (19%) 12

Total 111 36 147


The secondmost common destination of recycling in English is determiners. As we saw in example (1) above, speakers dosometimes produce a determiner and a noun and then recycle back to the determiner, or they may simply repeat thedeterminer before producing the noun, as in example (2). Both of these options stand in contrast to recycling back to thebeginning of the clause. Initial explorations of cases like (1) and (2) as opposed to examples like (6) suggest that recycling anentire clause may be associated with particular, typically disjoint, actions but further research is needed on the topic todetermine if this is indeed the case.

If we now consider the German data, given in Table 3 above, we see that subject pronouns, determiners, and prepositionsare about equal in frequency as destinations of recycling, at roughly 14–18% each. These 3 categories make up 62% of allfunction words serving as destinations of recycling; in fact, they make up half of all destinations of recycling. But no singlecategory stands out strongly as a primary destination of recycling. In what follows, we elaborate on the particularities ofGerman grammar resulting in these repair patterns.

First,wesee thatpersonalpronouns,whichwere themost frequently recycled syntactic category inEnglish, playamuch lesscrucial role in German (cf. also Rieger, 2003:58). In English, nearly half of all destinations of simple recycling were subjectpronouns (43%); in German, however, pronouns make up only 14% of simple recyclings. One possible explanation for thisdifference could be thedifferences inwordorder betweenEnglishandGerman. It is generally assumed that each languagehas abasic constituent order: For English it is generally SVO; for German it is either XVX (main clause), XXV (subordinate clause) orVXX (yes/no-question). And although the four-term case system in contemporary spoken German is subject to attrition andalthough there is a lot of syncretism in the system, case marking still allows the identification of grammatical relations (likesubject and object) independently of their position in the emerging turn. Thus we find not only variation in the position of thefinite verb, but also relatively free constituent order.4 This is especially true for the position of the so-called Vorfeld (front field,see footnote 5), which can be filled by almost any constituent. One option is of course a subject pronoun, as in example (12)below:

Table 3Syntactic category and repair type for German.

German Destination of recycling Replaced item Total

Function wordsSPronoun 14 (14%) 2 (5%) 16Determiner 14 (14%) 7 (16%) 16Preposition 18 (18%) 2 (5%) 20Connective 7 (7%) 0 (0%) 7Question-word 2 (2%) 1 (2%) 3Aux/Mod/Cop 2 (2%) 4 (9%) 6Adv + Det 7 (7%) 0 (0%) 7Other 10 (12%) 5 (15%) 7

Content wordsNoun/NP 3 (3%) 4 (9%) 7Verb 4 (4%) 12 (27%) 16Adjective 5 (5%) 6 (14%) 11Adverb 12 (12%) 2 (5%) 14

Total 98 44 138

4 Constituent order inGerman can be described by so-called ‘topological fields’ (cf. Drach, 1937): Front Field (FF), Left Sentence Bracket (LSB),Middle Field(MF), Right Sentence Bracket (RSB). The position of the finite verb varies depending on whether the sentence is a main or a subordinate clause. In mainclause order the finite verb fills the left sentence bracket and the front-field is filled. This type is called verb (finite)-second. In subordinate clause order thefinite verb fills the right sentence bracket and the left sentence bracket is filled with a subordinate conjunction: This type is called verb (finite)-last.

Main Clause:FF LSB MF RSBMorgen wird er kommenTomorrowwill he come

Subordinate Clause:FF LSB MF RSBØ weil er morgen kommen wird

because he tomorrow come will’The front field (Vorfeld), which is filled only in verb-second sentences (XVX), precedes the left sentence bracket and can contain any ‘‘Satzglied’’ (cf. FN5) of

the sentence. The left sentence bracket can be filled by the finite verb in verb-first (VXX) and verb-second sentences (XVX) or by a subordinating conjunction(complementizer e.g. dass, wenn, ob) in verb-last sentences (XXV). The right sentence bracket can be filled by non-finite verbal elements in verb-first andverb-second sentences with analytic tenses or it takes the whole verbal complex in verb-last sentences. The middle field (Mittelfeld) is defined as the partbetween the right and the left sentence bracket.


In example (12), the speaker repeats the subject pronoun ich (‘I’).But due to the relative freedom of word order in German, the front field position is frequently filled with adverbs.5 As we

saw in Table 3, adverbs serve as the destination of recycling in 12% of all instances.This observation leads to a generalization that puts adverbs together with ‘proper’ personal pronouns and a subset

of demonstrative pronouns. It seems to be the case that in German it is not the subject personal pronoun (as inEnglish) but the front-field, which serves as a common locus for recycling independently of syntactic category orgrammatical relation of the constituent that occurs there. But due to the significant variation in word order, there is nota high level of bonding between the front-field-constituent and the verb-second-position in German. This low level ofbonding also shows in the phonological realization. In contrast to English, there is no cliticization of auxiliary andcopula forms to pronouns in XVX utterances: *du’st (you’ve) is not a possible realization of du hast (you have) (cf. alsoRieger, 2003:64). While there are such cliticizations between certain verbs and some subject pronouns in interrogativesand in clauses in which a non-subject item occupies the front field (e.g. hasse for hast du, at least in the dialect ofGerman represented in our data), in these cases the verb precedes the subject clitic and thus they are not relevant to thepoint here, which is recycling back to subject pronouns. There is another observation that leads to a similar conclusion:English shows a fair number of recyclings back to a wh-word, while German shows very few such cases. Again we findcliticization between the wh-word and auxiliaries and the copula in English (who’s coming) but not in German (*wer’sgekommen?).

So a first step towards an answer to the question ‘‘Why is there less recycling back to subject pronouns in German?’’could be connected to the typological difference in word order between the two Germanic sister languages. German has amuch more flexible word order as compared to English and possibly due to that variability neither subject pronouns norwh-words have become hosts for auxiliary clitics as in English, and this early part of the turn has not become such a deeplyentrenched grammaticized unit (cf. Bybee, 2006) that would be rapidly available to speakers. German speakers wishing todelay production of an item early in the turn might therefore choose other options than recycling; for example, Rieger(2003) suggests that German speakersmake greater use of so-called filled pauses than do speakers of English. Future cross-linguistic research exploring all the options available to speakers for delaying next item due could shed light on thispossibility.

After pronouns, the next most common destinations of recycling in German are determiners and prepositions. AsTable 3 shows, these syntactic categories each make up 14–18% of all recyclings. We suggest that these are morecommon destinations of recycling in German than in English because German speakers tend to recycle back to thebeginning of a local syntactic constituent, such as the beginning of a noun phrase or prepositional phrase (Uhmann,1997), while English speakers have a greater tendency to recycle entire clauses (Fox et al., 1996; Fox and Wouk, 2003;Fox et al., to appear).

Leaving the class of function words and moving to an examination of content words, we notice that nouns,verbs, adjectives and adverbs do get recycled. But even when combined, they make up only 24% of the destinationsof recycling (remember that determiners and prepositions each reached 14–18%). Nouns are rarely recycled (3 instancesonly). Not a single instance was found in the German data where a noun was chosen as the destination of multi-wordrecycling. Verbs are similarly rare as the destinations of recycling: in fact, in each case, the verb was recycled byitself.

Turning now to Hebrew, above we saw that function words constitute 84% of all recycling destinations, whereascontent words constitute only 16%. In addition, 90% of all function words involved in self-repair constitute recyclingdestinations. The correlation ‘function-recycle’ is thus quite strong. If we examine Table 4 below, we see that prepositionand subject pronoun are the two main categories for destination of recycling, constituting 26% and 20%, respectively, ofall recyclings. Other syntactic categories contributing to recycling are discourse markers (10%), connectives (9%), andquestion words (8%).

As noted above, one of the most common functions of recycling has to do with delaying production of the next item due(cf. Fox et al., 1996; Rieger, 2003). This can be done for various reasons, both cognitive and social, such as in order to gainadditional cognitive planning time for the ensuing word or construction, to secure recipient gaze (Goodwin, 1981), topostpone a possible transition-relevance place (TRP, Sacks et al., 1974), etc.

Example (13) illustrates the second most common form of recycling in the Hebrew data, which is recycling back to asubject personal pronoun. The great majority of these recyclings are in a verbal clause:

5 The fact that adverbs can be placed in the front-field is one of the main features that serves as a test that they are indeed a proper ‘‘Satzglied’’, becauseonly constituents that can be topicalized (i.e. realized in the front field) do get that status.


Intonation unit 1 is a fragmentary intonation unit (Chafe, 1987, 1994), ending with a cutoff on the subject pronoun of theembedded clause. The speaker then begins a new intonation unit, repeating the subject pronoun but not the complementizerprocliticized to it.We see that, unlike the case of English, recycling can begin at a boundary between a clitic and theword it iscliticized to. Of course, in English clitics follow the words they are cliticized to (unless we follow Dixon (2007) in treatingmost subject pronouns, determiners and prepositions as clitics), while in Hebrew they tend to precede their hosts; so inutterances like (13) the speaker is recycling the host and not the clitic, something that English speakers do not have theopportunity to do.

The other environment inwhich speakers recycle back to a subject pronoun is in a nominal clause. Example (14) providesan illustration involving recycling of a demonstrative pronoun in a nominal clause:

Here we find vowel lengthening of the subject pronoun at line 1– another strategy for delaying next item due. Note thatMiri recycles the singular form of the demonstrative in this nominal clause, not maintaining agreement with the followingplural nominal predicate mishpatim (‘sentences’), in contrast to what we would expect according to Hebrew grammar.6

Table 4Distribution of repair type by syntactic category in Hebrew.

Destination of recycling Replaced item Total

Function wordsSubject pronoun 26 (20%) 3 (11%) 29Subject morpheme (bound on V) 1 (4%) 1Subject pronoun + V 3 (11%) 3Preposition 33 (26%) 3 (11%) 36Discourse marker 13 (10%) 1 (4%) 14Connective 12 (9%) 12Q-word 11 (9%) 2 (7%) 13Determiner 6 (5%) 1 (4%) 7Subordinator 4 (3%) 4Negative 3 (2%) 3

Content wordsNoun 5 (4%) 7 (26%) 12Verb 8 (6%) 3 (11%) 11CTP-V of saying 3 (2%) 3Verb + Prep 2 (7%) 2Adjective 2 (2%) 2Adverb 2 (2%) 2Other (content) 1 (4%) 1

Grand total 128 27

6 Intonation and context show that this is not a Hebrew cleft sentence.


Subject-predicate agreement is not alwaysmaintained in spoken discourse. Nevertheless, the pattern of recycling the subjectpronoun is common in these cases as well, suggesting that the delay here is for the purpose of delaying the content wordsrather than for agreement with the grammatical morphemes bound to them. Interestingly, in all cases of recycling back to asubject pronoun in a nominal clause, it is the unmarked masculine singular form of the demonstrative pronoun (ze) thatserves as destination of recycling.

In Modern Hebrew, generally an SVO language (Ravid, 1977), recycling the subject pronoun returns the speaker backto the beginning of the clause. By recycling back to this initial element, a speaker in part gains time for furtherprocessing of the ensuing clause, in agreement with Chafe’s Light Subject Constraint (1994:91). English subjects,according to Chafe, are typically ‘light’; i.e., not much linguistic material is employed in order to verbalize them, such asan unstressed pronoun. Hebrew is a so-called ‘Pro-drop’ language, which results in Hebrew subjects often being even‘lighter’ than they are in English, manifested only through a verbal suffix (and/or prefix, in the case of the future). Thisexplains the much lower rate of recycling back to the subject pronoun in Hebrew compared to the English rate of 43%.However, many sentences do begin with a subject pronoun, thus enabling subject pronouns to serve as the destinationof recycling in a fifth of all recycling cases (20%)—the second largest category serving as destination of recycling inHebrew.

For the largest syntactic category contributing to the high rate of Hebrew recycling, the preposition (26% of all recyclingdestinations), observe the following excerpt:

The Hebrew preposition b(e)- (‘in’), procliticized onto the noun (as are the common Hebrew prepositions: b(e)- (‘in’),l(e)- (‘to’), m(e)- (‘from’)), is lengthened, and then, following an intonation-unit internal pause, repeated preceding theverbalization of the noun toranut (‘duty’) in this prepositional phrase functioning as adverbial argument of the verb.These additional delaying phenomena provide further support that recycling here is related to delaying next contentword due.

In the case of a procliticized preposition preceding a definite noun, recycling the preposition entails recycling also thedeterminer fused with it, as in example (16):

The preposition b(e)- (‘in’) of line 4 fuses with the definite article ha- to form the proclitic ba- (‘in the’) procliticized ontothe noun. Both elements are then recycled prior to articulation of the noun kurs (‘course’). In Hebrew, by far the mostcommon determiner preceding the noun is the definite article.7 Since many definite articles are fused with the preposition

7 There is only one self-repaired determiner preceding the noun which is not a definite article in the entire corpus – the nonspecific pronoun ’eizeshehi(‘some’ fem. sg.).

There is no indefinite article in Hebrew.Hebrew demonstratives follow the noun, after repetition of the definite article, as in example 16, line 5:baku–rs haze.in the course the thisin this course.

Wesee that thedefinite article occurs twice in thisphrase: once fusedwith theprepositionbe- in the formofba-, the second timepreceding thedemonstrative ze.Hebrew possessive pronouns also follow the noun.


preceding them, as in example (16), this explains the fact that Hebrew determiners constitute only 5% of all recyclingdestinations.

Similar arguments about recycling back to a function word in order to delay next-item-due can be made for Hebrewdiscourse markers, connectives, subordinators, negatives, and question words.

The high rate of Hebrew recycling, then, is enabled by several syntactic and morphological facts in the grammar ofHebrew:

1) Modern Hebrew is generally an SVO language.

2) Hebrew grammatical elements generally precede the lexical elements they function as satellites to (unlike Japanese,e.g., cf. Fox et al., 1996). Subject pronouns, prepositions, discourse markers, connectives, determiners,8

subordinators, negatives, and question words, tend to appear preceding the verb, noun, phrase, clause, orconversational action (Ford and Thompson, 1996) they serve as satellites to, thus enabling the speaker to delayproduction of the lexical element while pausing on the grammatical one preceding it. As Fox et al. note, ‘‘lexicalitems are more contributionally consequential, semantically richer, and perhaps possibly less available (e.g., duringword searches) than the more restricted class of grammatical morphemes’’ (1996:232).

3) Apart from subject pronouns, Hebrew function words do not vary for person, number, or gender. They can berecycled while searching for the ensuing utterance, because they do not depend on its properties for their form(unlike the situation in German, see below).

We have seen that all three languages offer a resource for speakers that allows them to recycle a function word to delay aprojected content word. This is our first finding: languages that are genetically unrelated may show similar patterns ofrecycling and replacement if they are typologically similar.

3.2. Replacement strategies

Above we presented evidence that content words in each of the three languages are over-represented in replacementrepairs as compared to recycling repairs. We have argued that such differences arise from the different functions ofreplacement and recycling: replacement occurs when a speaker has produced part or all of a word or phrase that s/henow finds in some way incorrect or inappropriate; recycling is typically used to delay the next item due, which can beuseful in searching for a word (as well as for other purposes, such as requesting recipient gaze (Goodwin, 1979, 1981)).Given the much larger search space for content words, and the resulting larger opportunity for an incorrect orinappropriate choice, content words are more likely to be the target of a replacement than they are to be the destinationof recycling, at least in languages with modifiers before heads. Below we present detailed evidence from each languagedemonstrating this point.

From Table 2 we can see that for English, function words—subject pronouns, determiners, prepositions, wh-words,the existential item there, and connectives—make up a higher proportion of the destinations of recycling (85%),while content words—adverbs, nouns, adjectives, and verbs—make up a correspondingly lower proportion ofdestinations of recycling (15%), showing instead a higher presence in replacements (36%). While subject pronounsmake up 43% of all destinations of recycling, they constitute only 17% of all replacements. Nouns and verbsincrease in frequency in replacement versus recycling: nouns make up 4% of destinations of recycling and 19% of itemsreplaced (chi square is 11.2, p = 0.0008); verbs make up 4% of destinations of recycling and 11% of items replaced (chisquare = 3.9, p = 0.05). These patterns make sense given what we have said about the functions of recycling andreplacing.9

Nouns make up the largest category of replaced items (19%), with subject pronouns (which are function words) closebehind (17%), followed by verbs and determiners at 11%. The somewhat high rate of noun replacement seems to havetwo main sources: first, speakers’ initial production of a word is treated as infelicitous phonetically and is replaced witha different phonetic production; second, the speaker treats the initial attempt as interactionally inappropriate andreplaces it with another term from a related membership categorization collection (Sacks, 1992; Schegloff, 2007).Consider example (17) below; just prior to this fragment Jenn was recounting a recent event in which Bonnie asked amutual (male) friend if he was a ‘‘swinger’’; the friend, Timothy, responded by engaging in mock flirtation with Jenn. Inthe current interaction Bonnie expresses confusion about Timothy’s sexual orientation, and our fragment begins withBonnie offering one piece of information on the topic which appears to contradict the apparently heterosexual flirtationwith Jenn; at line 14 Jenn, in her third attempt at delicately formulating Timothy’s sexual orientation, replaces the boyswith men:

8 This does not include the definite article preceding an element modifying a definite noun, such as an adjective or demonstrative (such as the ha- (‘the’)preceding ze (‘this’) in the preceding footnote).

9 In English, we coded instances of replacing an entire NP (e.g. determiner + noun) as an instance of replacing a noun.


At line 5 Jenn produces a delicate formulation, with high pitch and facial expressions that could indicate that this wasexpected to be shared information, of Timothy’s sexual orientation. It is produced mostly in overlap, however, and at line 8Bonnie initiates repair. At line 9 Jenn repeats the formulation, thus treating the problem as one of hearing. However at line 11Bonnie initiates repair again, this time focusing on an understanding problem with the phrase for the boys. At line 14 Jennproduces the repair, shifting the verb to like; she first formulates the object of like as the boy- and replaces it with men.However for the boys might be used, it is possible that likes the boys could have a connotation of relationships with youngboys, as opposed to peers, and this interpretation could be treated as producing problematic inferences about Timothy. Boyexists in different collections of categories, including the stage-of-life collection (baby, boy, young man, man) and the sexcollection (boy, girl), and it may be this problematic ambiguity that Jenn seeks to remedy with the wordmen. The issue is nolonger treated as an understanding problem, and Bonnie goes on at line 16 to disagree with Jenn.10

While example (17) is the most dramatic instance of membership categorization repair we have in the English nounreplacement collection, the others show similar or related issues. By contrast, the verb replacement cases, which are muchsmaller in number, seem to have more to do with argument structure than careful formulation of membership, which ofcourse makes sense given the different grammatical and semantic work that nouns and verbs do. Perhaps this difference isone reason there are fewer verb replacements than noun replacements; another reason could be that English speakers preferto replace verbs with pre-framing recycling of the subject (given the strong bond between subject and verb; Helasvuo, 2001;Scheibman, 2002; Dixon, 2007), and such instances of pre-framing were not included in the current study.

Although verb replacements without pre-framing are quite uncommon in English, it may be worth examining aninstance. Consider example (18) below, in which the speaker replaces what appears to be the beginning of the verb instructwith the less specific verb do:

(18) Two Girls

1 Bee: y’know they(d) they do b- t!.hhhh they try even harder than a- y’know a regular instructor.

2 Ava: Righ [t.

3 Bee: [.hhhh to uh insr y’know do the class and everything.

Bee has just produced the word instructor at line 1 before she begins the grammatical increment at line 3; in fact, it ispossible that instructor (or regular instructor) is the outcome of a word search (initiated by the cut-off, and continued withy’know and a recycling of a) in a turn that has already involved several instances of repair; it is possible that this production ofinstructor is both the primer for the following verb and the reason the verb is not brought to completion and is replaced:speakers often avoid repeating words (or roots) in close succession.11

We turn now to German replacements. As we saw above, content words in German are over-represented in replacementrepairs. However, unlike English, there are two classes of function words that play an interesting role in replacement inGerman: auxiliaries/modals/copulas and determiners (9% and 16%, respectively, of all replaced items).

Although auxiliaries, modals and copulas are function words, they are far more likely to be replaced than to be thedestination of recycling. The reason for the high rate of auxiliary, copula and modal replacement in German appears to bethat speakers may find themselves having produced the auxiliary, modal or copula for one tense-aspect or number and thenfind that tense-aspect or number to be problematic. Consider example (19) below:

10 Suggesting of course that disagreement was in the air all along.11 In addition, a quick Google search on the phrase instruct the class reveals that in this collocation ‘‘instruct’’ is typically treated as a verb of speaking:instruct the class to do X, which is clearly not the sense Bee is after.


(19) dat sind* is jetz n BAUernHO:F

That are is now a farmhouse

‘that is now a farmhouse’

Determiners (articles, possessives, and demonstratives) get replaced in German for various reasons (demonstrativedeterminers are replaced by definite articles, definite articles are replaced by indefinite ones or singular forms are replacedby plural forms), and in the three instances reported on here the need for self-repair arises due to the gender classification ofGerman nouns, which is reflected in determiners. And as determiners precede their nouns, the gender-marked determinerscreate the need for repair and they are regularly replaced, if the noun – that is eventually produced – does not match thegendermarking on the determiner that has already been produced. Consider example (20) below, inwhich the speaker startswith a masculine article which is first recycled and then replaced by the feminine article, which is (retrospectively) thecorrect one for the noun Batterie (’battery’):

The following examples further illustrate the challenges faced by German speakers in selecting the appropriatedeterminer for anNP in the emerging turn. The challenge arises due to the gendermarking of nominal compounds, which canbe rather complex in German. And it is only the gender of the final noun (the grammatical head of the compound) thatdetermines the gender of the entire compound:

In extract (21) the compound Waffeditor (’wav editor’) is thus masculine because Editor is masculine:

The speaker, who has already produced a determiner marked for neuter gender (dieses), replaces it with the samedeterminer marked for masculine gender (dieser).

Turningnow to replacement of contentwords,we see that verbs are themost frequently replaced contentwords inGerman,and the rate of replacement is higher in German than in English (27% vs 11%), though not significantly so (chi square = 3.2,p = 0.07).WhyGerman speakers replace verbsmore often in our study is not clear, butwe expect it has to dowith a lack of pre-framedverbreplacement inXVX-sentences inGerman.WhileGermanspeakers frequentlyengage inverb replacementwithoutrecycling a prior term (Uhmann, 2006:192ff), because of the low bond between the front-field and the verb—which we sawabove in the low rate of pronoun recycling—English speakers are more likely to preframe a verb replacement, because of thetight bond between subject and verb in English. As a result, more verb replacements in English are excluded from the currentstudy than are verb replacements in German. However, as the numbers are small, the issue clearly requires further research.

Although nouns are not often replaced in German (and when they are, pre-framed recycling due to the tight bondbetween determiners and nouns because of gender marking is frequently involved here, and these instances are excludedfrom the current study), they do raise an interesting issue in German. Consider the following example:

In (22) the noun Gast (‘the guest’) has the correct article (der), because Gast is masculine, but although this turn isgrammatically correct, the speaker initiates repair and the entireNPgets replaced bydieGastkatze (‘the guest cat’) in the courseof the ongoing turn. In doing this kind of self-repair the speaker has to replace both the noun Gast by the compound Gastkatzeand themasculine determiner (der) with the female one (die), because the final part of the compound, Katze (‘cat’), is feminine.

In the next example the speaker replaces the determiner after the initiation of repair, but nevertheless ends up havingproduced a grammatically inappropriate article:

The speaker is engaged in producing a direct object. So in this instance the determiner is not only marked for gender butalso for case (accusative). The speaker comes up first with the term die gesamtvolumen, which shows a grammatically correctarticle, if Volumen is meant to be the plural (neuter, plural die). But obviously this is not the term he is aiming at and he nextproduces the compound Gesamtauftragsvolumen. This compound consists of three constituents and the last one is neuter(singular or plural): [N, neut. [Adj.gesamt] [N, masc. auftrag] [N, neut. volumen]]. The correct article (accusative) would thus be


either die (plural) or das (singular). The chosen article den only agrees with the masculine gender (accusative, singular) ofAuftrag (‘contract’), which is not the final noun and thus not the grammatical head of the compound. In this instance, then,the speaker starts off with what turns out to be the correct article and then repairs it with an incorrect article. Thus, self-repair involving nouns is regularly achieved through the replacement of the entire NP in German, since as determinersprecede their nouns and agree in gender and case, they have to be replaced, too. In our statistics, instances like (22) and (23)were counted only once as ‘replace NP’. If they were counted as well in the category ‘replace determiner’, the score would behigher in German than in English, which would reflect the differences in gender and case marking.

If we look at replacements in Hebrew, we see that the syntactic category contributingmost to replacements is noun (26%of all replaced items). Next come the categories of subject pronoun, subject pronoun + verb, and preposition (eachcontributing 11%). There are no instances of replaced adjectives or adverbs.

Excerpt (24) is an example of replacing a noun.

Here the speaker confused the identity of the responder in the narrated story, and as a consequence replaced the wrongreference, the beginning of aba (‘father’), with the right one – the proper name Shaxar (line 3). Again we see the pattern ofemploying the strategy of replacement when a speaker has produced part or all of a word or phrase that s/he now finds insome way incorrect or inappropriate. Here, as in all cases of noun replacement, the replacing and replaced items are of thesame syntactic category.

Subjectpronouns contribute11%of all replacements andare replaced for reasons similar to thoseofnouns—incorrectness orinappropriateness of the replaced pronoun. However, the replacing item is not always of the same syntactic category. In thefollowing excerpt, for instance, the subject pronoun is replaced by a noun, most likely in order to make the referent clearer:

The pronoun ze (‘it’) is replaced with the noun phrase hamaxma’a hazot (‘this compliment’), probably because it is judgedtoo ambiguous by the speaker in this context.

Hebrew replaces verbs at the same rate as English: (11%). However, if we add to this category the V + Prep category (thereason for doing sowould be because replacement of certain verbsmay result in replacement of the preposition they require,and so the replacement of the preposition is obligatory here), the rate goes up to 18%. This rate is lower than in German,though not significantly so. Hebrew verbs include an obligatory bound morpheme agreeing with the subject, sometimes inaddition to an overt subject, sometimes without an overt subject (as in the case of so-called ‘‘Pro-drop’ languages).Technically, this morpheme gets recycled in all cases of replacing the verb.12 However, because the subject morpheme is anobligatory morpheme on every Hebrew verb, these cases are classified here as replacements rather than as ‘other’.

Verb replacement is illustrated in example (26), concerning a second grade teacher who went through every singlestudent of the class in order to determine whether or not to punish him or her:

12 In the past and present tenses, this implies postrecycling, since the subject morpheme consists of a suffix. In the future, the subject morpheme consistsof both a prefix and a suffix.


The intransitive verb ’amda (‘stood’) is replaced by the causative verb he’emida (‘made stand’), with recycling of theroot’.m.d. (relating to the concept of standing) and the obligatory 3rd person feminine singular suffix -a. Note that the 3rdperson pronoun hi (‘she’) preceding ’amda (line 3) does not get repeated. The source of this repair is a lexical choice error onthe part of the speaker, which was repaired by replacing the problematic item.

Another category contributing 11% of the cases to the Replace category involves replacement of the subject pronoun alongwith the verb. Examine the following excerpt:

In this sequence of two consecutive instances of self-repair, the speaker first replaces the plural subject pronoun hem(‘they’) with the singular one, hu (‘he’), resulting in replacement of the verb form as well (although its stem remains thesame), so that ba’u (‘came’ 3rd PL) becomes ba (‘came’ 3rd SG). She then replaces the singular subject pronoun + verb backwith the plural one, resulting in her original choice. In English, the verbwould simply be post-recycled here, but in a languagelike Hebrew, with richer verb morphology, the result is a case of replacement of two items.

There is one set of verbs that is never replaced in our database, and that is the complement-taking predicate (CTP,Thompson, 2002) verb-of-saying, preceding constructed dialogue (Tannen, 1989). As seen in Table 4, none of these verbs arereplaced in the corpus – they always serve as recycling destinations. There are 3 self-repaired CTP verbs in the corpus, all ofthem involving verbs of saying in either 1st or 2nd person, thus always referring to discourse participantswhich are typicallynot overtly expressed in spoken Hebrew (Polak-Yitzhaki, 2004). This, then, is a sub-group of the cases involving no overtpersonal pronoun, which therefore serve as recycling destinations, e.g.:


In the absence of a subject pronoun here, we see that the speaker recycles back to the CTP verb-of-saying siparti (‘told’, 1stSG).

In summary, we have found that noun is the category most often replaced in English and Hebrew, while verb plays thatrole in German. In German, ‘helping’ verbs constitute about 1/5 of all replacements, while English shows amuch lower rate ofreplacement for those verbs. We have explained these particular patterns of replacement repair with reference to themorpho-syntactic patterns of each language.

4. Comparing English, Hebrew and German

We are now in a position to discuss the central finding of the study, which is that, while all three languages show atendency (1) to recycle back to function words, and (2) to replace content words at a relatively high rate, each languageshows a unique constellation of syntactic constituents involved in recyclings and replacement repairs.

The first major difference we noted among the languages was the prominence of subject pronoun as a destination ofrecycling. In English, subject pronoun is an extremely common destination of recycling, making up 43% of all destinations ofrecycling; in German and Hebrew, however, the rate is much lower (14% for German and 20% for Hebrew). We believe thereason for this difference is the very high rate of clause-initial overt subject pronouns in English—nearly every clause inEnglish has an overt, clause-initial subject pronoun. German and Hebrew, on the other hand, showmuch greater variabilityin the placement of subject pronouns and, especially for Hebrew, in whether subject pronouns are overtly expressed or not.In German the more important category of destination of recycling is front field, which may be occupied by a subjectpronoun, but may also be occupied by other parts of speech, such as adverbs, non-finite verbs, non-subject NPs, and so on. Inaddition, in German there is only amorphological bond (agreement) and no phonological bond between subject pronouns infront field and their verbs—there is no cliticization of verbs to their subjects in front field, as there is in English. Thus subjectpronoun is not a heavily entrenched clause-initial locus in German.WhileHebrew speakers do tend to recycle back to subjectpronouns, the common practice of leaving subjects unexpressed often produces utterances in which the verb is initial, andthus verbs are a more suitable destination of recycling in Hebrew than in English.

The second difference we noted was the pattern of replacement regarding determiners. In German, determiners aremarked for the number, case and gender of the noun they accompany, and thus if a determiner with one set of suffixes isproduced, projecting one type of noun, and the speaker changes the noun before producing it, then, if the new noun differs innumber and/or gender from the projected noun, the speaker will tend to replace the determiner to match the new noun.English and Hebrew speakers do not engage in this practice, since prenominal determiners in those languages do not changemorphology depending on number/gender/case of the noun.

The third differencewe focused onwas the higher rate of noun replacement in English and Hebrew, and the higher rate ofverb replacement in German. We suggested that the lower rate of verb replacement in English is due at least in part to thetendency of English speakers to recycle the subject pronoun before replacing a verb, resulting in a small number of instancesof simple verb replacement; German and Hebrew speakers tend not to have this tendency to preframe verb replacements asstrongly as English speakers. The low rate of simple noun replacement in German may result from excluding pre-framingrecycling from this study.

The fourth difference we noted was the different frequencies of recycling back to determiners and prepositions in thethree languages. While all three languages have determiners and prepositions, their prominence in recycling varies. InEnglish and German, prepositions are a fairly common destination of recycling, but in Hebrew prepositions are the mostcommon destination of recycling, at almost double the rate in English and German. We suggested that this is due to the factthat prepositions in Hebrew tend to be proclitics and thus speakers must recycle back to them. On the other hand, in EnglishandGerman determiners tend to be recycled back to at a rate of 12–14%, while the rate in Hebrew is lower (5%– a statisticallysignificant difference from German; chi square = 6.34; p = 0.01). We have seen that Hebrew determiners often fuse with theprepositions preceding them, resulting in the recycling going all theway back to the preposition. This contributes both to thelow rate of recycling back to determiners and to the high rate of recycling back to prepositions in Hebrew.

Although there are striking differences across the three languages, there are also striking similarities. All three languagesshow a strong preference for functionwords as destinations of recycling in recycling repairs, and all three languages show anoverrepresentation of content words in replacements (although this overrepresentation is much stronger in German than inEnglish or Hebrew). This patternmakes sense given the fact that all three languages tend to have functionwords that precedecontentwords: prepositions, determiners, pronouns, and auxiliaries andmodals all13 tend to precede the contentwords theyserve as satellites to in these languages (althoughHebrew does not have auxiliaries ormodals in the same sense as in Englishand German). We have seen that function words can be recycled to delay next content-word-due, and thus are likely to be

13 Word order of Hebrew NPs is actually more complex.Whereas prepositions and the definite article precede the noun; demonstratives, possessives, andadjectives (along with the definite articles preceding them in an NP) follow the noun, e.g.:babayit hagadol haze sheliin the house the big the this my‘in this big house of mine’

However, these noun satellites which follow the noun are very seldom repaired in our data.In addition, word order in German VPs is more complex. Auxiliaries and modals precede their full verbs only in verb-second and verb-first sentences

with an empty middle field. In verb-last position they follow their verbs.


used as the destinations of recycling. In contrast, content words can be inapposite and thus need to be replaced, leading to ahigher rate of replacement among content words than recycling.

Of course in English and Hebrew functionwords do sometimes pose troubles for speakers (e.g. an inapposite preposition,or an interactional pressure to switch from I towe to include the recipient—see Lerner and Kitzinger, 2007), and thusmay bereplaced; in German, as noted above, the need for agreement in gender, case and number between determiners and nounsmay lead speakers to replace determiners if the projected noun due is changed.

The importance of word order in our findings cannot be overstated. In fact, if we look at a language with different wordorder characteristics, the strong association between function words and recycling appears not to hold. Consider, forexample, Korean. Korean is a classic example of a language in which modifiers follow their heads; for example adpositionsare actually suffixes on their nouns. While Korean has demonstrative determiners, they are much less frequently used thanare articles in English, German and Hebrew. Subjects are only infrequently expressed overtly, and pronouns in general arenot common. Thus Korean has few function words that could serve to delay the next item due (see Fox et al., 1996, for thispoint regarding Japanese).Wewould thus expect that functionwordswould not show a special associationwith recycling inKorean. And indeed that’s what we find. In a cursory exploration of 159 instances of self-repair in Korean, collected by HyunJung Yang, we found that only 25% of simple recyclings had as their destination of recycling a function word, compared to85% for English, 84% for Hebrew, and 78% for German (see Table 1).

The associations between part of speech and type of repair that we found in Hebrew, English and German thus reflect thetypological characteristics of those three languages and are not to be taken as a universal fact about languages (cf. Fox et al.,in press).

5. Conclusions

This study has explored the recycling and replacement patterns in three languages, two ofwhich are sister languages. Thisstudy supports the findings of prior work on self-repair (cf. Gomez de Garcia, 1994; Fox et al., 1996; Fincke, 1999; Uhmann,2001, 2006; Rieger, 2003; Wouk, 2005) which suggest that typological characteristics of a language, such as majorconstituent order, presence or absence of determiners which precede or follow their nouns, presence or absence of overtarguments (especially subjects), presence or absence of highly entrenched—even cliticized—relationships between certainfunction and contentwords, or between certain functionwords (as in the case of Hebrew prepositions and determiners), andpresence or absence of complex morphology on function and/or content words, all shape the self-repair practices of thatlanguage.

This relationship between typological characteristics and repair practices further suggests that even genetically closelanguages may exhibit quite different repair practices if their structures have diverged substantially. This is clearly the casefor English and German, which, though sister languages, show distinctly different repair practices. One of the mostremarkablemanifestations of this difference is the treatment of subject pronouns in recycling in the two languages: althoughboth languages show a strong preference for overt subject arguments (a typologically rare feature), the greater flexibility inword order in German leads away from the entrenchment of subject as the beginning of a clause, and especially leads awayfrom the entrenchment of the linear subject-verb bond. In English the deep entrenchment of the subject-verb bond can beseen in the cliticization of certain high-frequency verbs (auxiliaries and the copula be) on subjects. While certain subjectpronouns can cliticize to a few verbs in German, these clitics follow their verbs and thus do not increase the frequency ofrecycling back to subject.

In a similar vein, languages which are not genetically close but which share certain typological features may exhibitsimilar repair practices. German and Hebrew, for example, though not genetically related, both exhibit a much greatertendency to recycle back to prepositions than does English. We have argued that this pattern arises from the very closemorphological and phonological relationships between prepositions and the determiners and nouns that follow them inGerman andHebrew: both languages showan extremely close phonological relationship between someprepositions and thefollowing nouns or determiners; in Hebrew some prepositions fuse with the immediately following definite article andprocliticize to their nouns, and in German some prepositions have fused with certain determiners to make a single word.Moreover, in German prepositions require their noun phrases to be in specific morphological cases, so there is also a closemorphological relationship in German between prepositions and their noun phrases. No such phonological ormorphologicalrelationships exist in English between prepositions and their noun phrases.

In addition to correlating self-repair strategies with language typology, through the study of self-repair, we gain anunderstanding of the way the speakers of a language themselves interpret the structure of their language. Self-repairpatterns reveal the degree of connection that speakers create among particular syntactic categories. For instance, self-repairpatterns involving frequent recycling back to a clitic or to a clitic host, such as to a preposition procliticized to a Hebrewnounor to a pronoun to which the copula or auxiliary is encliticized in English, reveal speakers’ understanding of morphemes andthe connections among them in their languages. Such a window proves to be important for our understanding of languageorganization, as the recent interest in the concept of ‘word’, and in particular in the ambivalent nature of cliticswith regard to‘words’, suggests (see Dixon and Aikhenvald, 2002; Dixon, 2007).

Furthermore, it seems that clitics shape self-repair practices in crucial ways in our three languages. For example, the factthat a variety of verbal elements (the copula, auxiliaries, some modals) cliticize to subject pronouns (and to full NP subjectsas well) in English helps to create the importance of the subject as a destination of recycling. And in turn the importance of


the subject as a destination of recycling helps to create turn-beginning in English as a critical destination of recycling (nottrue of all languages; see Fox et al., 1996). InHebrew the fact thatmany prepositions are procliticized to their nouns leads to ahigh rate of recycling of prepositions, and in German subject pronoun clitics follow their verbs, and thus do not increase thefrequency of recycling back to subject.

In addition to these differences among the three languages, there is a strong similarity, namely that all three languageshave functionwordswhich precede the contentwords they serve as adjuncts to. Aswe saw above, this similaritymanifests inthe tendency in all three languages for speakers to recycle back to functionwords rather than content words. Our prediction,on the basis of the data presented here, is that all languages with functionwords that precede their respective contentwords(which will tend to be verb-initial and verb-medial languages) will show a preference for recycling back to function wordsrather than content words, while languages without such function words will not show this preference (cf. Fox et al., inpreparation). Further research on languages of various typological profiles will be needed to determine if this prediction isaccurate.

Acknowledgements

This researchwas supported by grant BCS0406512 from the National Science Foundation.We are grateful to the audienceat the 2006 ICCA-Conference in Helsinki for the rare chance to not only present an earlier version of the paper twice but foralso having two discussions. We are also grateful to Fay Wouk for comments on an earlier version, and to Vince Sarich forhelp with statistics on an earlier draft. Some of the English data were coded by Alexander Ferguson, who was funded by agrant from the Undergraduate Research Opportunity Program at the University of Colorado, Boulder. We are also grateful toMareike Stausberg and Regina Pustet, who helped to code the German data. Yael Maschler would like to thank the FreiburgInstitute for Advanced Studies (FRIAS) at Albert-Ludwigs-Universitat, Freiburg, Germany, for an External Senior Fellowshipduring which the manuscript was revised. We also acknowledge the insightful comments of three anonymous reviewers.

References

Bybee, Joan, 2006. From usage to grammar: the mind’s response to repetition. Language 82, 711–733.Chafe,Wallace, 1987. Cognitive constraints on information flow. In: Tomlin, R.S. (Ed.), Typological Studies in Language, vol. 11: Coherence and Grounding in

Discourse. John Benjamins, Amsterdam, pp. 21–51.Chafe, Wallace, 1994. Discourse, Consciousness, and Time: The Flow and Displacement of Conscious Experience in Speaking and Writing. University of

Chicago Press, Chicago.Clark, Herbert, Wasow, Thomas, 1998. Repeating words in spontaneous speech. Cognitive Psychology 37, 201–242.Dixon, R.M.W., 2007. Clitics in English. English Studies 88, 574–600.Dixon, R.M.W., Aikhenvald, Alexandra (Eds.), 2002. Word: A Cross-Linguistic Typology. Cambridge University Press, Cambridge.Drach, Erich, 1937. Grundgedanken der Deutschen Satzlehre. Frankfurt.Fincke, Steven, 1999. The syntactic organization of repair in Bikol. In: Fox, B., Jurafsky, D., Michaelis, L. (Eds.), Cognition and Function in Language. CSLI,

Stanford, pp. 252–267.Ford, Cecilia E., Thompson, Sandra A., 1996. Interactional units in conversation: syntactic, intonational and pragmatic resources for the management of

turns. In: Ochs, E., Schegloff, E.A., Thompson, S. (Eds.), Interaction and Grammar. Cambridge University Press, Cambridge, England, pp. 134–184.Fox, Barbara, Hayashi, Makoto, Jasperson, Robert, 1996. Resources and repair. In: Ochs, E., Schegloff, E., Thompson, S. (Eds.), Interaction and Grammar.

Cambridge University Press, Cambridge, pp. 185–237.Fox, Barbara, Jasperson, Robert, 1995. A syntactic exploration of repair in English conversation. In: Davis, P. (Ed.), Descriptive and Theoretical Modes in the

Alternative Linguistics. John Benjamins, Amsterdam, pp. 77–134.Fox, Barbara, Wouk, Fay, 2003. Proposal to National Science Foundation. .Fox, Barbara, Wouk, Fay, Hayashi, Makoto, Fincke, Steven, Tao, Liang, Sorjonen, Marja-Leena, Laakso, Minna, Wilfrido, Flores Hernandez, 2009. A cross-

linguistic investigation of the site of initiation in same-turn self-repair. In: Sidnell, Jack (Ed.), Comparative Studies in Conversation Analysis. CambridgeUniversity Press, Cambridge.

Fox, B., Wouk, F. et al. To appear. A Cross-Linguistics Study of Self-Repair. Amsterdam: John Benjamins.Goodwin, Charles, 1979. The interactive construction of a sentence in natural conversation. In: Psathas, G. (Ed.), Everyday Language: Studies in

Ethnomethodology. Irvington Press, New York, pp. 97–121.Goodwin, C., 1981. Conversational Organization: Interaction Between Speakers and Hearers. Academic Press, New York.Gomez de Garcia, Jule, 1994. Communicative Strategies in Conversational Kickapoo. PhD Dissertation, University of Colorado, Boulder.Hayashi, Makoto, 1994. A comparative study of self-repair in English and Japanese conversation. In: Akatsuka, N. (Ed.), Japanese/Korean Linguistics IV. CSLI,

Stanford, pp. 77–93.Helasvuo, M., 2001. Syntax in the Making: The Emergence of Syntactic Units in Finnish Conversation. Benjamins, Amsterdam.Hockett, Charles, 1967. Where the tongue slips, there slip I. In: To Honor Roman Jakobson, Mouton, The Hague, pp. 910–936.Jesperson, Otto, 1924. The Philosophy of Grammar. The University of Chicago Press, Chicago.Karkkainen, Elise, Sorjonen, Marja-Leena, Helasvuo, Marja-Liisa, 2007. Discourse structure. In: Shopen, T. (Ed.), Language Typology and Syntactic

Description. second edition. Cambridge University Press, Cambridge, pp. 301–371.Labov, William, 1972. Language in the Inner City. University of Pennsylvania Press, Philadelphia.Lerner, Gene, Kitzinger, Celia, 2007. Extraction and aggregation in the repair of individual and collective self-reference. Discourse Studies 9, 526–557.Levelt, Willem, 1982. Zelfcorrecties in het spreekproces. Mededelingen der Koninklijke Nederlandse Akademie van Wetenschappen 45, 215–228.Levelt, Willem, 1983. Monitoring and self-repair in speech. Cognition 14, 41–104.Lickley, R.J., 1994. Detecting Disfluency in Spontaneous Speech. Doctoral Dissertation, University of Edinburgh.Maclay, H., Osgood, Charles, 1959. Hesitation phenomena in spontaneous English speech. Word 15, 19–44.Maschler, Yael, 2009. The Haifa Corpus of Spoken Israeli Hebrew. http://hevra.haifa.ac.il/com/maschler/.Oh, Sun-Young, 2005. English zero anaphora as an interactional resource. Research on Language and Social Interaction 38, 267–302.Polak-Yitzhaki, Hilla, 2004. tafkideyhem shel kinuyey haguf haprudim, haxavurim vehaproklitiyim betafkid nose taxbiri basiax ha’ivri hadavur (‘The

functions of subject personal pronouns (attached, overt and proclitic) in Spoken Israeli Hebrew’). Unpublished M.A. Thesis, University of Haifa (inHebrew).

Ravid, Dorit, 1977. mispar heibetim shel be’ayat seder hamarkivim be’ivrit yisra’elit modernit (‘Several aspects of the problem of element order in ModernIsraeli Hebrew’). balshanut ‘ivrit xofshit (‘Hebrew Computational, Formal, and Applied Linguistics’) 11, 1–45 (in Hebrew).


https://ssl.haifa.ac.il/com/maschler/,DanaInfo=hevra.haifa.ac.il+

Rieger, Caroline, 2003. Repetitions as self-repair strategies in English and German conversations. Journal of Pragmatics 35, 47–69.Sacks, H., 1992. Lectures on Conversation. Blackwell, Mass.Sacks, Harvey, Schegloff, Emanuel, Jefferson, Gail, 1974. A simplest systematics for the organization of turn-taking for conversation. Language 50, 696–735.Schegloff, Emanuel, Jefferson, Gail, Sacks, Harvey, 1977. The preference for self-correction in the organization of repair in conversation. Language 53, 361–

382.Schegloff, Emanuel, 1979. The relevance of repair to syntax-for-conversation. In: Givon, T. (Ed.), Syntax and Semantics, vol. 12. Academic Press, New York,

pp. 261–286.Schegloff, Emanuel, 1987. Recycled turn beginnings. In: Button, G., Lee, J.R.E. (Eds.), Talk and Social Organization. Multilingual Matters, Clevedon, England,

pp. 70–85.Schegloff, E., 2007. Sequence Organization in Interaction: A Primer in Conversation Analysis. Cambridge University Press, Cambridge.Scheibman, J., 2002. Point of view and grammar: Structural patterns and subjectivity in American English conversation. John Benjamins, Amsterdam.Tannen, Deborah, 1989. Talking Voices: Repetition, Dialogue, and Imagery in Conversational Discourse. Cambridge University Press, Cambridge.Thompson, Sandra, 2002. ‘‘Object complements’’ and conversation: towards a realistic account. Studies in Language 26 (1), 125–164.Uhmann, Susanne, 1997. Selbstreparaturen in Alltagsdialogen. Ein Fall fur eine integrative Konversationstheorie. In: Schlobinski, P. (Ed.), Syntax des

gesprochenen Deutsch. Westdeutscher Verlag, Opladen, pp. 157–180.Uhmann, Susanne, 2001. Some arguments for the relevance of syntax to same-sentence self-repair in everyday German conversation. In: Selting, M.,Couper-

Kuhlen, E. (Eds.), Studies in Interactional Linguistics. John Benjamins, Amsterdam, pp. 373–404.Uhmann, Susanne, 2006. Grammatik und Interaktion: form follows function?—function follows form?. In: Deppermann, A., et al. (Eds.), Grammatik und

Interaktion. Untersuchungen zumZusammenhang von grammatischen Strukturen undGesprachsprozessen. Verlag fur Gesprachsforschung, Radolfzell,pp. 179–202.

Wouk, Fay., 2005. The syntax of repair in Indonesian. Discourse Studies 7, 237–258.Yang, Hyun Jung, 2003. Self-repair in Korean. MS.

Barbara A. Fox is Professor of Linguistics at the University of Colorado, Boulder. Her interests center on grammar in its reflexive relationship to interaction, as wellas on grammar in use more generally. She also works on philosophies of language-in-use.

Yael Maschler teaches linguistics at the Department of Communication and at the Department of Hebrew Language, University of Haifa, Israel. Her researchfocuses on the grammaticization of discourse patterns into linguistic structure. She has published articles on how grammar is both constitutive of and emergentfrom interaction in the fields of bilingual discourse, discourse markers, syntax of spoken language, and stance-taking. Her book Metalanguage in Interaction:Hebrew Discourse Markers was recently published by John Benjamins.

Susanne Uhmann is Professor of Linguistics in the German Department at the University of Wuppertal (Germany). Her main research interest focuses oninvestigating the interdependency between grammar and interaction observed in naturally-occurring everyday conversations. Her theoretical background is inlinguistics (syntax and phonology) and conversation analysis, and this is reflected in her major publication Grammatische Regeln und konversationelle Strategien(1997). She works chiefly on syntactic structures of contemporary spoken German, self-repair, prosody (with special focus on intonation, rhythm, and speechrate), as well as information structure.


Fox, Barbara, Maschler, Yael, and Uhmann, Susanne. (2010). A cross-linguistic study of self-repair:...

Documents

Transcript of Fox, Barbara, Maschler, Yael, and Uhmann, Susanne. (2010). A cross-linguistic study of self-repair:...