Dialogue & Discourse 15(2) (2024) 85-112 doi: 10.5210/dad.2024.203 ©2024 Dagnew Mache Asgede This is an open-access article distributed under the terms of a Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0/). Self-Repair in Tigrinya: Trouble Sources, Mechanisms and Solutions Dagnew Mache Asgede dagnewmache.a@gmail.com Department of English Language & Literature https://orcid.org/0000-0002-1109-4415 Arba Minch University Arba Minch, Ethiopia Editor: Kallirroi Georgila Submitted 11/2022; Accepted 07/2024; Published online 10/2024 Abstract This paper analyzes conversational self-repair, which refers to reconstructing problematic portions of a prior oral discourse by oneself, in Tigrinya. Tigrinya is a Northern Ethio-Eritrean-Semitic language spoken by the inhabitants of the Tigray regional state of Ethiopia and Eritrea. This article relies on recorded oral data from speakers of the Rayya Tigrinya variety, particularly inhabitants of Neksege located to the West of Maichew. A conversational analysis (CA) approach is used to analyze the trouble sources, mechanisms (initiators), and results of self-repair. The article shows that pronunciation problems emanating from dialectal variation or tongue slip, wrong word order including focus misplacement, missing constituents, perceived misunderstandings, and using (totally) wrong constituents are some of the trouble sources that push speakers to repair portions of a prior oral utterance. On top of that, cut-offs, particles, and lexemes (one verbal noun and some predicates) are identified as self-repair initiators. Though cut-offs do not indicate a self-repair, particles and predicates may sometimes indicate a self-repair. Finally, the article posits that expanding, replacing, re-ordering, aborting and restarting, and inserting are some of the solutions set for the repairable segments in the repaired portions of the oral discourse. The author recommends for further investigation repair in the process of language acquisition and learning, and the relationship between self-repair and the demographic features of participants. Keywords: Tigrinya, self-repair, trouble sources, repair mechanisms, repair solutions 1 Introduction Linguistic repair refers to addressing any language-related concerns during the development of oral discourse. Repair can be self-initiated or other-initiated. This paper specifically focuses on self-initiated repair, which involves correcting one’s speech to improve communication. Various scholars have studied self-repair in different languages. For example, Fox et al. (1996) state that the grammatical structure of a language influences how speakers correct their speech. Murphy (2019) also indicates a positive relationship between repairs and the phonological and morpho-syntactic features of languages. Repair, in general and self-repair in particular, is one of the most overlooked areas of linguistic research in Tigrinya, and in Ethiopian languages as a whole. This paper aims to investigate the sources of trouble, mechanisms, and solutions in Tigrinya to contribute to the discussion on conversational self-repair. The work is based on annotated oral data. The paper is divided into four main sections. The first section, literature, introduces the sociolinguistic and grammatical features of Tigrinya for readers who are unfamiliar with the language, as well as basic concepts of conversational repair and different types of repairs. The second section covers methods and data, providing details on the data collection tools used, information about the data, protocols for annotating the data, and how inter-rater agreement was calculated among four raters. The third section presents and analyzes the data, focusing on trouble sources, repair mechanisms, and the results of self-repair. Finally, a discussion is provided based on the data results. https://orcid.org/0000-0002-1109-4415 ASGEDE 86 2 Literature 2.1 The Tigrinya Tigrinya, spoken by the people of the Tigray Regional State in Ethiopia and by Eritreans, is one of the Semitic language branches of the Afro-Asiatic language phylum. Tigrinya, along with Geez and Tigre, are classified as a northern Ethio-Eritrean-Semitic language group. Though the language is said to have several dialects, no single study has attempted to delineate the dialect numbers based on linguistic characteristics; researchers ignored the dialectal mutual intelligibility of the language’s varieties. Roughly speaking, we can assume that the language has several dialects; of these, the one spoken by the people of Tigray Regional State’s southern zone, where the data this article relies on is recorded from, is known as the Rayya Tigrinya variety. The language is used as a medium of instruction from kindergarten to grade eight, and it is taught as a course in grades nine through twelve. The language is also taught at Abiyi Adi Teachers’ College and Mekelle University at diploma and degree levels respectively. At the moment that I am writing this article, due to the devastating war on Tigray regional state, none of the mentioned programs are active. Tigrinya has 37 consonants (Asgede, 2019). Asgede (2019) and Mehari (2021) argue that the consonant segments /p/, /p’/, /v/ and /ʒ/ are ‘non-existent’ in Tigrinya except in loan words. Asgede (2019) claims that loan words contain sounds that are difficult to pronounce for the illiterate part of the speech community. Concerning vowel sounds, the language has seven common vowels; the majority of the speakers of Rayya and Wajerat dialects do not utter the middle front unrounded vowel /e/ which is common in the quasi- standard Tigrinya variety (Asgede, 2019; Mehari, 2021). Tigrinya has CV and CVC syllabic structure (Yohannes, 2002; Asgede, 2019; Mehari, 2021). In word- initial and final, a consonant cluster is impermissible; if there is either a word-initial or final potential consonant cluster, the epenthetic vowel (high central unrounded) /-ɨ-/ and the high front unrounded vowel /-i/ are inserted respectively. When two vowels appear in sequence in complex words that contain two or more morphs, the semivowels /-w-/ or /-j-/ or the glottal consonant /-ʔ-/ are inserted as epenthetic consonants. Besides, consonant gemination that is phonemic in most cases is common in Tigrinya at word medial but never in other word positions. Tigrinya is an inflectional (i.e., fusional under the synthetic morphological type) language where several morphs with different grammatical meanings appear in a single word, and a single morph encodes several meanings and/or grammatical functions simultaneously. It is morphologically complex to the extent that a word contains elements that are equal to a sentence. This makes the gloss and translation of the data challenging, which leads to some tolerable problems. For example, the roots of the verbs in the language are templates of consonants into which vowels are inserted, and affixes are attached to mark different grammatical functions. The word order for Tigrinya simple declarative sentence is S-O-V as in ʔɨssu goggo bəlɨʕu ‘He ate enjera’. However, this does not work when all grammatical elements are indicated in a single word as in the following example. (1) Introspective Data bəlɨʕ-u-ww-o bəlɨʕ-u-ww-o eat:PRV-SJ:3SM-Ø-OBJ:3SM ‘He ate it (M).’ The order of the affixes that mark the subject and object follows the verb. The language has also circumfix that marks negation as in ʔajtɨχədɨj [ʔaj-tɨ-χəd-ɨ-j ‘NEG-2SMːSUB-goːIPV-Ø-NEG’] ‘you will not go’. Though the grammatical elements of the language are relatively emphasized in previous works, the features of conversation have been overlooked by researchers. As far as my reading is concerned, it is only Asgede (2023; 2019) who tried to touch on some of the conversational features (particularly the performance-related elements) of Tigrinya. This article therefore will focus on describing self-repair as one SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 87 feature of conversation. To achieve this, the paper describes the trouble sources, the mechanisms, and the results of self-repair in different genres of oral usage of the language. In addition to a grammatical description, Teferra (1979) and Blejer (1986) provided some brief discussions on some pragmatic elements like discourse markers (DMs) of the language. They described particles, conjunctions, adverbials, and interjections as the main sources of DMs. Teferra (1979) has noted that =s and/or =si ‘as for’ and ʔɨmma or its clitic form =mma ‘focus marker’ are focus marking particles; =ja ‘evidential marker’ an enclitic, ʔɨkko or its clitic form =kko ‘change in focus’, dəʔa ‘and so, then’; ʔɨmmo ‘then’; wəy dɨmma ‘or’. Blejer (1986) also stated that the enclitic =s and/or =si ‘as for’ does mark focus. Though previous studies like Mehari (2021; 2011), Asgede (2019; 2007), Girmay (2012), Nigusse (2012), Yohannes (2002), and Kiros (2009) have specifically tried to deal with different linguistic features of the different Tigrinya varieties, it is Asgede (2023; 2019) who gives room (though limited) to the conversational features of the Rayya language variety. 2.2 Conversation Repair Conversational repair refers to pointing back to a problematic segment of oral discourse and pointing forward to its more accurate replacement. In other words, repair refers to a process of reformulating a proposition. It is one of the performance matters of conversation or speech; this process may be initiated by cut-offs, filled pauses, lexical constituents, etc. Filled pauses are linguistic forms that help a speaker and listener get more time to deal with problematic information that is given in a portion of a prior segment (Brennan and Schober, 2001 cited in Hlavac, 2011). As to Deese (1984) and Brennan and Williams (1995), filled pauses draw the attention of a listener to pay due emphasis to what follows in the development of speech. Roughly speaking, after Fox et al. (1996), a repair can be defined as a conversational process “by which speakers correct errors they have made in their immediately prior talk”. However, a repair takes place even in the absence of any error committed; repairable segments may be repaired for clarification, rephrasing, and explanations. With this, Schegloff et al. (1977) stated that a repair is not limited to error correction, but a repair is made even if there are no errors made. In this paper, a repair is used as a conversational process that allows speakers to reconsider their ‘problematic’ prior utterances. Speakers repair a portion of their prior constituents in two ways: self-initiated and other-initiated whose differences are briefly explained in the following sub-section. 2.3 Types of Repairs A repair can be either other-initiated or self-initiated. In a conversation, when something that a speaker has said is misheard or when a speaker encounters a problem in understanding a previous utterance, an attempt is often made to elicit a repetition, clarification, elaboration, or correction, referred to in Conversation Analysis as other-initiation of repair (Schegloff et al., 1977 cited in Ha and Grice, 2017). Other-initiated repair refers to either correcting others’ prior utterance or being corrected by others. A self-repair in contrast refers to reconsidering your portion of prior discourse; it refers to repairing your problematic prior utterance (self-initiation). A self-repair is made by speakers when they recognize they had made a mistake or their prior discourse is problematic in any way. Self-repair may be accomplished within an oral discourse unit like sentence and turn, or even after long strings of constituents and turns are spoken out. Self-initiated repair within the same turn may be initiated by varieties of non-lexical speech perturbations such as cut-offs and repetitions, and sound stretches (Schegloff et al., 1977). Similarly, depending on the mini-corpus that this study relies on, in Tigrinya, repair takes place either after part of a word and sometimes even after an entire word or after a long discourse segment is uttered. For instance, as Asgede (2019) mentions, reformulating is dominantly signalled by cut-off words, as can be seen in example (2) below. ASGEDE 88 (2) (Taken from Asgede (2019: 215)) s’aʕɨda ʃɨngurti ʔɨm... d͡ʒɨ.. rɨħus d͡ʒɨnd͡ʒɨbɨl rɨħus d͡ʒɨnd͡ʒɨbɨl ʔabaʕɨχə səssəg s’aʕɨd ʃɨngurt ʔɨm... d͡ʒɨ.. rɨħus d͡ʒɨnd͡ʒɨbɨl rɨħu d͡ʒɨnd͡ʒɨbɨl ʔabaʕɨχə səssəg white onion DM_PLNPRC DM_CNVRPR wet ginger wet ginger fenugreek basil ‘Garlic, wet ginger, fenugreek, basil…’ In example (2), the speaker started to utter the word d͡ʒɨnd͡ʒɨbɨl ‘ginger’ but remembered that the word should be modified by the adjective rɨħus ‘wet’. By implication, the ginger needs to be wet, not dry. It seems that the speaker remembers she had missed a constituent that should describe the partially uttered word; then, she cuts off the word to add an adjective that describes the exact state of the ginger she is referring to. The cut-offs that are considered here are those that initiate repairing (or might have other purposes) but are followed by the corrected/revised version of the repairing source. The current article focuses on such self-initiated repairing trouble sources, repair mechanisms, and repair results. 3 Methods and Data 3.1 Methods and Procedures of Data Collection 3.1.1 Tools of Data Collection Data was gathered by recording informal interviews, spontaneous conversations, and tales. This was accomplished during a three-month personal field trip, from April 10th to July 17th 2019, in Neksege (located in the Southern Tigray Administrative Zone). The recorded data was processed using software Audacity, and the format was exported as a wav file. The wav files containing the oral data were imported into ELAN, which was used to segment each text into discourse units such as turns, external interventions, sentence and clause boundaries. The segmented data was phonetically transcribed; morphs of each complex word were marked; an English gloss was assigned to each morph, and each discourse unit received a literal and free translation in English. 3.1.2 Data Statement The mini-corpus, temporarily named as Rayya Tigrinya mini-speech corpus, was established during a three- month fieldwork in which the researcher recorded the speech data. The speech corpus was established from recordings of spontaneous conversations, conflict resolutions, informal interviews, descriptions of how to do things, biographies, and tales. Originally, the mini-corpus totalled 24 hours of recorded audio/video data. For research ethics purposes, based on the basic schema recommended by Bender and Friedman (2018), the data statement is detailed as follows. The data was annotated by four annotators who were asked for their judgement on the sources of troubles, self-repairing mechanisms, and self-repair solutions of the 400 identified self-repair markers (for a detailed discussion on the data statement please see Appendix II). 3.1.3 Annotation Protocol There are established terminologies used in the literature in relation to how to establish structure in the annotation for repair. It is clear that a self-repair is initiated by an original utterance (Hough, 2015) that I prefer to call trouble source, which is named as reparandum by Meteer et al. (1995) and Shriberg (1994). This type of constituent is split into two, namely start and reparandum by Ginzburg et al. (2007). The trouble source of an utterance that underwent repair is defined as “the entire stretch of speech to be deleted” by Shriberg (1994). That constituent is preceded and followed by square brackets [] (as in [constituent]). The other element that needs clarification is the repair mechanisms that are termed as interregnum which refers to an editing phase marked by “filled pauses or editing terms” (Meteer et al., 1995; Shriberg, 1994; Hough, 2015). This section is termed as editing term (Ginzburg et al., 2007), editing phase (Levelt, 1983), cut-off to repair (Nakatani and Hirschberg, 1994) and repair initiation (Schegloff, 1992). This part of a constituent is preceded and followed by {} as in {constituent} and in a bold font. SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 89 The third concept that is relevant to the present article is repair solutions that are named differently in the literature: repair (Hough, 2015; Shriberg, 1994; Meteer et al., 1995), repair proper (Schegloff, 1992), and alteration (Ginzburg et al., 2007). This is marked by # as in #constituent#. This description is explained as follows. In the literature, an interruption point (Meteer et al., 1995; Shriberg, 1994) named as moment of interruption by Levelt (1983), Blackmer and Mitton (1991), and Ginzburg et al. (2007), that is the moment that appears between the trouble source and repair mechanism, is not considered in the analysis for it demands a different experiment and may not be understood by considering the surface structure of a discourse. Not only that but some authors like Hough (2015) and Ginzburg et al. (2007) mentioned that there is a portion of discourse called continuation that is not considered here for I believe it does not affect the nature of the self-repair. 3.1.4 Inter-Annotator Agreement Based on an example text that contained 15 repairs, the researcher trained the annotators and raters for half a day. The content of the training was how to identify the repair typologies and how to annotate (transcribe) them. Based on Healey et al. (2005), a short guideline encompassing some detailed points is given to the annotators. For details on the guideline and the mean of the identified self-repair markers, see appendix III. Considering the total number of repairs identified by the researcher (see Appendix II) and the remaining three annotators, the mean is 390. Therefore, the 400 repairs are accepted as valued and reliable. This is substantiated with the inter-rater agreement among the four raters (including the author) for the categories of self-repairs: trouble sources, repair mechanisms, and results of self-repairs under sections 4.1., 4.2., and 4.3., respectively. The Fleiss’ kappa model is employed to manually compute the inter-rater agreement for there are four raters (more than two raters) and the data is nominal (Fleiss, 1971; 2003). 𝐾 = Po−Pe 1−Pe where K is Fleiss’ kappa; Po is observed agreement, Pe is expected agreement among the raters The inter-rater agreement for the sources of troubles in self-repair, repair mechanisms, and results of self-repairs is calculated in the analysis section using the Fleiss’ kappa interpretation intervals provided in the equation above. 3.2 Data Analysis Framework The data analysis framework used is Conversation Analysis (CA), which is a method for studying human social interaction in the field of linguistics (Stivers and Sidnell, 2013). According to Maynard (2013), CA has “established itself as a worldwide theoretical and empirical endeavor concerned with the social scientific understanding and analysis of [human social] interaction”. CA necessitates distinct data derived from spontaneous and naturally occurring social interaction (Maynard, 2013). Maynard (2013) adds that this approach is concerned with how language is used in “social, publicly interpretable methods and behaviors”. CA uses audio and/or video recordings of naturally occurring communicative events like an everyday conversation, instructions/explanations of how to do things, ritual speeches, story tales, and stories of people’s lives to study the detailed actions of participants. The audio-video data is significant because sɨʔləj ʔɨbba daɲɲəw ʔɨkko məs’u ʔɨnho [sɨʔləj] {ʔɨbba} #daɲɲəw# ʔɨkko məs’u ʔɨnho trouble source repair mechanisms repair solutions ‘Si’eley I mean Dagnew has come.’ ASGEDE 90 CA aims to “describe the organization of ordinary social activities such as taking turns at talking” (Mondanda, 2013). Repair is a common phenomenon in conversation. CA aids in the investigation of how and why repairs are performed. According to Yilmaz (2004), a repair can be initiated by oneself or by someone else. This feature of conversation indicates either disfluencies, difficulties activating language units and/or concepts, hearing, and understanding. Justifying why a speaker requires repair may be difficult because the reasons in most cases are cognitive and remain unknown. Taking an inquisitive look at both the linguistic and social contexts, as well as looking into the meanings and functions of repair mechanisms in the text, can help to uncover the hidden cognitive reasons a speaker performed a repair (Schourup, 1985). In this regard, CA aided the researcher in examining the corpus neutral of hypothesis or theory; this approach helps to record theory neutral linguistic data. This is what Wong and Waring (2010) refer to as “unmotivated looking”. To make using the CA approach easier, the entire corpus was transcribed. Whenever possible, the transcription included “not only what was said, but also how it was said” (Wooffitt, 2005) by providing literal and free meanings of extractions. Similar to the work of many researchers for other languages (Jabeen et al., 2011; Ward, 1998; Kawamori et al., 1998; Heeman et al., 1999; de Rooji, 2000; Matras, 2000; Bell, 2010; Wang, 2011), here CA is used as a method to analyze the linguistic features and pragmatic functions of trouble sources such as discourse markers, filled pauses, cut-offs, etc. in Tigrinya. Based on the host context, the function and meaning of each identified self-repair marker were considered. Examining what happened to the repair source and its alteration text, it is attempted to investigate repair techniques, reasons for repairing, and devices used to indicate that repairing is being initiated. 4 Data Presentation and Analysis 4.1 Trouble Sources of Self-Repairing Self-repairing occurs in a conversation for a reason; trouble sources cause a speaker to reconsider a portion of an old utterance. Self-initiated repair is triggered by any of the following conversational issues: diction problems caused by dialect variations and social relationships of the participants, pronunciation errors (and tongue slip), word order issues, missing constituents, focus misplacement, and misunderstandings. The distribution of the problems of the repairs identified from the data is described in the table below for clarity. Table 1 demonstrates that the majority of the repair initiators are pronunciation errors and tongue slips (90 items), followed by diction problems (68 items), missing constituents (65 items), word order problems (49 items), and misunderstandings (43 items). As demonstrated in Table 2 below, the inter-rater agreement Trouble/Repair Initiator Types Total Number of Cases Identified Percentage Diction problems Dialect Variation Participants Social Relationship 68 32 36 17 47.059 52.941 Pronunciation Errors and Tongue Slips 90 22.5 Word Order Problems 49 12.25 Missing Constituents 65 16.25 Focus Misplacements 33 8.25 Misunderstandings 43 10.75 (Totally) Wrong Constituents 51 12.75 Unidentified 1 0.25 Total 400 100% Table 1. List of trouble source types. SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 91 D ic ti o n P ro b le m s P ro n . E rr o rs an d T o n g u e S li p s W o rd O rd er P ro b le m s M is si n g C o n st it u en ts F o cu s M is p la ce m en ts M is u n d er st an d in g s W ro n g C o n st it u en ts U n id en ti fi ed S u m Sum 277 349 199 255 141 169 200 10 1600 P 0.173 0.218 0.124 0.159 0.088 0.106 0.125 0.006 Po 0.86 Pe 0.153 K 0.835 Table 2. Inter-rater agreement on the trouble sources. among the four raters on the trouble sources is computed using the Fleiss’ kappa model that is represented with the formula mentioned in section 3.1.4 and in Appendix III. Based on the suggested Fleiss’ kappa interpretation (see Table 10 in Appendix III), the inter-rater agreement is 0.835, which is ‘almost perfect’. 4.1.1 Pronunciation Errors Though not universal, and certainly not consistent across data collected from a speech community of a specific language or variety, the data available reveals that the pronunciation problem and tongue slip, which account for 22.5% of the trouble sources of the repairs identified, are the most powerful motivators for speakers to make a conversational repair. (3) ARG1 səlus səlus [lajəs’a] #nab jəs’a# ʕɨdəga [tɨχəd] {ʔa...} #tɨχəjɨd# malət ʔɨju səlus səlus la=jəs’a Na b jəs’a ʕɨdəga tɨ-χəd ʔa… Tuesday Tuesday to-Yetsa To Yetsa market 2SM:SBJ-go:IPV part tɨ-χəjɨd malət ʔɨj-u 2SM:SBJ-go:IPV means COP-3SM ‘Every Tuesday, you go to Yetsa for Market.’ The proclitic la= ‘to’ in lajəs’a ‘to Yetsa’ functions similarly to the preposition nab ‘to’ in grammatical and syntactic terms. The speaker has repaired the lajəs’a, which was a source of trouble during the repair, and replaced it with nab jəs’a to correct the way the prepositional phrase should be pronounced. The allative marker in quasi-standard Tigrinya is the preposition nab ‘to’, but it is commonly attached to the next noun by Rayya Tigrinya variety speakers who habitat in Neksege; by other Rayya Tigrinya variety speaker communities, it is realized as da= ‘to’ as in daməxələ ‘to Mekele’. In the same turn, the speaker corrected the way he pronounced the word tɨχəd as tɨχəjɨd ‘you go’. The problematic and corrected constituents are similar in meaning but differ in pronunciation due to dialectal variation; the source of the repair therefore seems to be a pronunciation problem that arose from dialectical differences between the participants. 4.1.2 Diction Problems In addition to ‘inappropriate’ pronunciation, diction-related trouble is a repair initiator. Diction, at least in this article, is influenced by the participants’ social relationships as well as dialect variations. This issue accounts for 68 (17%) of the total 400 repairs identified in the mini-corpus. ASGEDE 92 (4) ASG2 ʔatta ʔabzi labəlaʕɨnadə ʔabzi tɨħar.. tɨftɨnwa rɨħɨχ’ ʋəl ʔatt-a ʔab-ʔɨz-i la-bəlaʕ-ɨna-də ʔabzi tɨħar.. VOC-2SM LOC-PRX-2SM PROG-eat:IPV-SJ:1PL-Q LOC-PRX-2SM RPR tɨ-ftɨn-wa rɨħɨχ’ bəl 2SMː2SBJ-urinateːIPV-FOC away:CN say:IMP:2SM ‘Please urinate away for we are feeding ourselves here.’ The repair initiators are the cut-off constituents of the above extraction; the speaker was to utter tɨħarɨʔ ‘urinate’, which the speech community interprets as derogatory. Instead, the speaker used the repair result constituent tɨftɨn ‘urinate’, which is a meaning extended from ‘trying, experimenting’. The repair result constituent tɨftɨn ‘urinate’ is a commonly used word by the quasi-standard Tigrinya speakers and is used in media and written texts across Tigray. The speaker seems to believe that the trouble source constituent is ‘inappropriate’ in the context of people eating food. Besides, words may be replaced by others as a result of a repair process arising from the participants’ social relationships. (5) AST1-3 [ħɨləf] #ħalɨfom# kəf jɨʋəlu ħɨləf ħalɨf-om kəf jɨ-bəl-u pass:IMP:SJ:2SM pass:IMP-2SM.H sit JUSS-say:IMP-2SM.H ‘Pass through…please be seated in the next seat.’ QSH1-1 ʔɨʃʃɨni ʔɨzi wəddəj marjam taχɨbɨrχa ʔɨʃʃɨ-ni ʔɨz-i wəddi-əj marjam tɨ-ʔa-χɨbɨr-χa okay-PLT PRX-2SM son-RL:1S (Saint) Merry JUSS:SBJ:3SF-CAUS:3SF-respect:IPV-OBJ:2SM ‘Okay, my sonǃ May Saint Merry respect you.’ AST1-4 ʔamjən #jɨk’ɨrta# dəjalələχukum ʔɨkko jɨʔu ʔatta dɨʋəχukum ʔamjən jɨk’ɨrta daj-a-lələj-χu-kum ʔɨkko jɨʔ-u Amen sorry NEG-CAUS-notice:CN-SJ:1S-OBJ:2SM.H FOC COP-3SM ʔatt-a dɨ-bəl-χu-kum VOC-2SM REL-say:PRV-SJ:1S-OBJ:2SM.H Amen! I’m sorry for I called you as if you are my peer for I didn’t notice it was you. QSH1-2 daħna ʔɨzi wəddəj ʔarɨmχajjakko daħna ʔɨz-i wəddi-əj ʔarɨm-χa-jj-a-kko fine PRX-2SM son-RL:1S correct:PRV-SJ:2SM-Ø-OBJ:3SF-FOC ‘My son, do not mention it; you already corrected it (the inappropriate speech).’ As depicted in example (5), ħɨləf ‘(2SM) pass through’ is an imperative mood that is not respectful. The repair result which is ħalɨfom ‘after (2SM.H) pass through’ has an honorific marker in it. Even though there is no covered verbal repair initiator, it is understood that the speaker replaces the ħɨləf with ħalɨfom because the speaker recognizes the addressee is a priest who has a high social class at least from a religious perspective. This becomes clear when the main speaker apologizes which is explicitly communicated by the constituent jɨk’ɨrta ‘sorry’ (AST1-4) and the other speaker accepts and even acknowledges the interlocutor unknowingly spoke ‘inappropriately’. 4.1.3 Missing Constituent Spontaneous speech has many disfluencies which differentiate spoken genres from written genres. Omitting a constituent is one of the disfluencies. Constituents of a single discourse unit (say for instance a turn) and long constituents (a discourse unit like sentences) might be omitted. The latter refers to missing, for example, a step or steps while describing how to do things or missing part of a plot when narrating stories and tales. This type of conversational feature is another trouble source that leads speakers to self-repair a repairable prior segment. SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 93 (6) BMA4-51 [ʔɨta ʔom] #ʔɨta nəwaħ ʔom# ʔɨmma hɨzχajja ħɨləf laʕaddɨmma hɨzχajja ħɨləf ʔɨt-a ʔom ʔɨta nəwaħ ʔom ʔɨmma hɨz-χa-jj-a DEST-3SF Tree DEST-3SF long:SF tree FOC hold:IPV-SJ:2SM-Ø-OBJ:3SF ħɨləf la-ʕaddi-mma hɨz-χa-jj-a ħɨləf pass:IMP:SJ:2SM ALL-home-FOC hold:IPV-SJ:2SM-Ø-OBJ:3SF pass:IMP:SJ:2SM ‘You bring the tree… the long tree to home.’ DMA4-29 /la.. tʃ’əllə la.. tʃ’əllə RPR okay ‘Okay’ Example (6) shows how missing constituents in the same turn initiate repair. The main speaker wants his interlocutor to bring a tree and later he remembered the tree should be not just the tree but the long tree than the other alternative trees. The tree, therefore, needs to be described by the adjective nəwaħ ‘long:SF’. This is the linguistic context that tells the participants that there are two or more trees, and the main speaker is saying that the interlocutor should bring the long tree with the repaired constituent. In the same turn, the main speaker spoke to the addressee to bring the tree to a place that was not mentioned originally in the trouble source; the other participant was also to ask a question and started uttering the word labəj ‘to where’ but cut-off as la… However, before the interlocutor uttered his question (which seems to not have been heard by the main speaker), the main speaker noticed he had forgotten to mention the destination and repaired the problematic constituent as laʕadimma hɨzχaja ħɨləf ‘bring it to home:FOC’. 4.1.4 Wrong Constituent Commonly, speakers utter wrong constituents in their speech. Soon, they repair their speech and replace the wrongly uttered constituent with an appropriate one. (7) KDW1-7 [sɨʔləj] {ʔɨbba} #daɲɲəw# ʔɨkko məs’u ʔɨnho sɨʔləj ʔɨbba daɲɲəw ʔɨkko məs’-u ʔɨnh-o NAME FOC\RPR NAME FOC come:PRV-SJ:3SM exist:IPV-SJ:3SM ‘Si’eley I mean Dagnew has come.’ In the above excerpt, the speaker wanted to talk about Dagnew but he spoke out Si’eley who is the small brother of Dagnew. Here, the repair is initiated by a trouble source of uttering a wrong constituent. Concerning this, speakers also put constituents in their jumbled order. To fix the wrong constituent, a speaker repairs his or her speech by himself or herself. 4.1.5 Wrong Constituents Order As can be seen in the following example (8), in the first turn of the extraction, the speaker uttered bota nɨfas dɨnhəwwo ‘place windy’ but later he noted that the constituents are in their wrong order, and the adjectival phrase nɨfaj dɨnhəwwo ‘that has wind; windy’ should precede the head noun bota ‘place’ that is the intended constituent to be modified by the modifier. The repair is initiated due to the wrongly ordered constituents. (8) BMA4-22 məd͡ʒəmərja [bota nɨfas dɨnhəwwo] {ʔɨm} #nɨfas dɨnhəwwo bota# tɨmərɨs’ məd͡ʒəmərja bota nɨfas dɨnhəwwo ʔɨm nɨfas first place wind REL-exist:IPV-SJ:3SM-OBJ:3SM RPR wind dɨ-ʔɨnh-ə-ww-o bota tɨ-mərɨs’ REL-exist:IPV-SJ:3SM-OBJ:3SM place IMP:2SM-select:IPV ‘First, place, you select a windy (specific) place.’ DMA4-17 ʔɨʃʃi ʔɨʃʃi Okay ‘I am listening to you; please proceed to talk.’ ASGEDE 94 BMA4-23 kaʔu [sɨfɨħ ʔablɨχa tɨs’ərɨg bota] {malət} #sɨfɨħ ʔablɨχa bota tɨs’ərɨg# kaʔu sɨfɨħ ʔa-bl-ɨχa tɨ-s’ərɨg bota malət sɨfɨħ then wide CAUS-do:IPV-SBJ:2SM IMP:2SM-clean place mean wide ʔablɨχa bota tɨ-s’ərɨg CAUS-do:IPV-SBJ:2SM place IMP:2SM-clean ‘Then, you clean a place wide I mean you clean a wide area.’ In the third turn (BMA4-23) of the above excerpt, see how the wrongly ordered constituents are the trouble source of the repair which is initiated by the verbal noun malət ‘I mean; means’. The noun bota ‘place’ follows the verb tɨs’ərɨg ‘you clean’ (the verb preceded the object) which violates the word order of the language that is S-O-V. Wrongly ordered constituents refer not only to words but also to long constituents of a discourse. For instance, as can be seen in example (9), the wrongly ordered constituents are two clauses that contain one step each. (9) MQE3-16 gəβɨt͡ ʃə tɨt͡ ʃ’awətləχa [məd͡ʒəmərta ʔɨmni tarsɨn kaʔu ʔab χɨltə tɨbʷadən] {ʔɨbba} #məd͡ʒəməja ʔab χɨltə tɨbʷadən kaʔu ʔɨmni tarsɨn# gəβɨt͡ ʃə tɨ-t͡ ʃ’awət-ʔɨntələ-χa məd͡ʒəmərta ʔɨmni t-a-rsɨn kaʔu ʔab gebiche IPR-play-when-SBJ:2SM first stone IPV-CAU:2-heat then into χɨltə tɨ-bʷadən ʔɨbba məd͡ʒəməja ʔab χɨltə tɨbʷadən kaʔu ʔɨmni two IPV-group:2SM I mean first into two IPV-group:2SM then stone t-a-rsɨn IPV-CAU:2-heat ‘When you play gebiche (an indigenous game), you first heat stone; then, form two teams; I mean, first, form two teams and heat a stone.’ The first bold constituent in the above excerpt contains two steps that should be completed in a serial order before boys play an indigenous game called gebiche; however, the speaker has wrongly first uttered the second step, which is the trouble source of the repair that aims at reconsidering the way the speaker thought the ideas should appear. The second bold portion of the excerpt thus is the corrected order of the steps which were jumbled in the first part of the excerpt. 4.1.6 Focus Misplacement Though scarcely, focus misplacement is also observed as another type of trouble that motivates speakers to repair their portion of the prior utterance. (10) AGS1-22 [ʔɨssu səbʔaj] {ʔɨk..} #ʔɨssu ʔɨkko səbʔaj# jɨʔu ʔɨssu səbʔaj ʔɨk.. ʔɨssu ʔɨkko səbʔaj jɨʔ-u 3SM man RPR 3SM FOC Man COP-3SM ‘He man.. He is a man’ The cut-off initiates the repair process. In the repairable segment, the focus was meant in the səbʔaj ‘man’, but the speaker wants to shift the focus from səbʔaj ‘man’ to ʔɨssu ‘he’ through the process of the repair mechanism. 4.1.7 Perceived Misunderstanding When a current speaker who controls a particular turn believes that a listener may not understand, or may be confused by the information he/she has just heard, the speaker may repair. Perceived misunderstanding is another cause of repair during a conversation. Misunderstanding is a trouble source that accounts for 10.75% of the total repair sources among the 400 repairs identified. SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 95 (11) WRA1-7 [ʔanəs χalɨʔ wulɨd jəbləj] ʔanə-s χalɨʔ wulɨd jə-bl-ə-j 1S-FOC another child NEG-say:IPV-RL:1S-NEG ‘I do not have another child.’ (lit. He is the only son that I have.) WRA1-8 #mɨwuladɨskka kʷuwankʷa jɨʔuwa# mɨwulad-ɨs-kka kʷuwankʷa jɨʔ-u-wa giving birth-FOC-FOC language COP-3SM-FOC ‘Blood relationship is meaningless.’ (lit. One might have a good child who is not a child by blood.) LMA1-11 ʔɨwwə ħak’iχi ʔɨwwə ħak’i-χi yes true-POSS:2SF ‘Yes, what you said is true.’ WRA1-9 #ʔɨgzihar jɨməsgən ʔɨsχatɨχum ʔɨnhəχumni# {malətəj} jɨʔu ʔɨgzihar jɨ-məsgən ʔɨsχa-tɨχum ʔɨnhə-χum-ni malət-əj God JUSS:3SM-thanks 2SM-PL exist:IPV-SJ:2PL-OBJ:1S means-POSS:1S jɨʔ-u COP-3SM ‘Thanks to God, you are here for me.’ WRA1-10 hamʔu gɨn ʔab məjda wədɨχ’ə χ’arjə ham-ʔu gɨn ʔab məjda wədɨχ’-ə χ’arj-ə like-3SM but in plain lay:CN-1S remain:PRV-1S ‘He have left me alone.’ (lit. As to him, however, I am thrown around the field.) The excerpt above demonstrates two steps of repair: the main speaker explains why the speaker’s only son, who lives in Jigjiga, did not visit her. She stated in the segment WRA1-7 that she has no other biological child besides him. She is aware, however, that the listener has acknowledged and assisted her as if she is his biological mother. As a result, the main speaker perceived that the listener might be uncomfortable with what she had said. This perceived misunderstanding motivates her to reframe her speech. The second constituent of the discourse is a softening repair to the previous constituent of the utterance, though there is no explicit repair initiator. When we look at the turn WRA1-9, it has a clear repair initiator that is malətəj ‘I mean’. This begins as the main speaker expands on the previous turn by providing additional clarification. As a result, perceived misunderstanding is another issue that drives a speaker to make a self-initiated conversational repair. 4.2 Repair Mechanisms The dominant possible repairing initiations which I call repairing mechanisms markers are the following: cut-offs, filled pauses, particles, predicates, pauses, and non-verbal such as facial expressions and gestures. For details on the repair mechanisms, look at the following table (Table 3). Type repairing mechanism Number Percentage Cut-offs 238 59.5 Filled Pauses 60 15 Particles 38 9.5 Predicates 32 8 Pauses 21 5.25 Visible only (like Facial Expressions and Gestures) 9 2.25 Unidentified 2 0.5 Total 400 100% Table 3. Types of repair mechanisms. ASGEDE 96 C u t- o ff s F il le d P au se s P ar ti cl es P re d ic at es P au se s F ac ia l E x p re ss io n s an d G es tu re s U n id en ti fi ed S u m Sum 948 238 148 128 86 36 16 1600 P 0.593 0.149 0.093 0.08 0.054 0.023 0.01 Pe 0.392 Po 0.961 K 0.936 Table 4. Inter-rater agreement on the repair mechanisms. As demonstrated in Table 4, the computed inter-rater agreement is 0.936, which is interpreted as almost perfect agreement (see Table 10 in Appendix III). To gain an understanding of the repair initiating markers in Tigrinya, let us have a further discussion on each of the techniques with some examples. 4.2.1 Cut-offs Note that all cut-offs may not necessarily have relations with a repair; for example they may mark hesitations. In the mini corpus, at least 181 cut-offs were found not to have an association with a repair. As roughly examined, most of these 181 cut-offs are associated with the hesitation and/or planning process (see Asgede, 2019). (12) DL2-79 {ʔɨto..} #ʔɨsχatɨχum naχ’adəm təmharo sɨlə dɨχonχum gobəzat jəχum# ʔɨto.. ʔɨsχatɨχum na-χ’adəm təmhari-o sɨlə dɨ-χon-χum gobəz-at RPR 2PLM POSS-past time student:PL for REL-become-2PL clever-PL jəχ-um being-2PL ‘RPR Since you are from the old batches, you are clever students.’ ʔɨto... is part of an unfinished constituent of the host utterance of the main speaker, as shown in example (12). The speaker then begins to utter a different word, indicating that the unfinished word was either inappropriate or incorrect. The process is more of a repair than a hesitation. Distinguishing between repair and hesitation is difficult; when a speaker resumes an utterance with a completely different constituent than the one that was left unfinished, the researcher refers to it as a repair. However, hesitation occurs when the speaker restarts her/his utterance with a similar part of the unfinished part of the language unit. (13) HLF3-33 [ʔɨti wannaʔu] {ʔɨk..} #ʔɨti wanna k’umnəgərukko# ʕarsɨχa mɨχʔal jɨʔu ʔɨti wanna-ʔ-u ʔɨk.. ʔɨti Wanna k’umnəgər-u=kko DEF:3SGM main-Ø-DEF RPR DEF:3SGM Main meaningful-DEF:3SGM=FOC ʕarsɨ-χa mɨχʔal jɨʔ-u self-2SGM help COP-3SGM ‘The main RPR.. the main thing is to be to able help yourself.’ You can see that there is no possible similarity between ʔɨk.. (the unfinished language unit in example (13) above) and the immediately following word of the same segment. The speaker had recognized k’umnəgəru ‘the main thing’ shall receive a focus that is marked by =kko that the speaker misplaced. It is for this reason that the speaker cuts off and re-utters as can be seen in example (13). SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 97 (14) ASG1-7 [ʔɨziʔa] {ħa..} #ʔɨziʔa dɨbəynas# sələste s’ɨmdi tawʕɨlɨla malət jɨʔu ʔɨz-a-ʔ-a ħa.. ʔɨz-a-ʔ-a dɨ-bəyn-a-s sələste PRX-3SF-Ø-SNG:3SF RPR PRX-3SF-Ø-SNG:3SF for-alone-3sf-DM_FOC three s’ɨmdi ta-wʕɨl-ʔɨl-a malət jɨʔ-u pair CAUS-spend the day-exist:nonpast-3SF Means COP-3SM ‘This (pointing at a farming land) consumes three pairs.’ The speaker has uttered the unfinished word (ħa..) that later appeared in his speech as dɨbəynas ‘alone:F’ which is the second word after the repair is made. In this particular example, the speaker replaced the repair mechanism with a different constituent. (15) GDY1-102 ʔɨzaw mɨsza bɨmərfɨʔ gəβrɨna nəgga... nɨləgba malət jɨʔu ʔɨz-a-w mɨs-ʔɨz-a bɨmərfɨʔ gəβr-ɨna {nəgga..} PRX-3SF-SNG with-PRX-3SF INST-needle use:IPR-1PL RPR #nɨ-ləgb-a# malət jɨʔ-u SJ:3PL-fasten-OBJ:3SF Means COP-3SM ‘It means we sew this (pointing at a cloth on his left hand) with this one (pointing at a cloth in front of him) using a needle.’ DMA1-37 ʔɨh ʔɨh PART ‘I am listening; please proceed’ GDY1-103 kaʔu bək’k’a tɨməlɨsχa tɨbl.. tɨχdəna kaʔu bək’k’a tɨ-məlɨs-χa {tɨbl..} #tɨ-χdən-a# then Just REF-back-2SM RPR REF-wear:IPR-OBJ:3SF ‘Then, you just wear it.’ The necessity of repairing may also be for language style or word diction as in example (15). The speaker prefers nɨləgba over nəga.. (nəgannanɨja) which are synonyms. For diction purposes, the speaker prefers to select a word over another word, but this came to the mind of the speaker after the less preferred word started to come out of the speaker’s mouth. The speaker repairs the original speech to put the more appropriate word in the sentence. 4.2.2 Filled Pauses Filled pauses, also called fillers, are linguistic units that are “used instead of pauses or facial expressions” (Asgede, 2019). In terms of function, fillers “shed light on the speech production process” and are “indicative of the mental processes underlying speech generation” (Swerts, 1998). In spontaneous interactions, filled pauses signal many other functions, as the speaker is going to repair an original part of the oral discourse. Though it isn’t easy to discriminate whether fillers serve different functions like planning process, hesitation, and repair (Hlavac, 2011; Brennan and Williams, 1995; Asgede, 2019), this is noticeable when we look at the repaired portion of the discourse. (16) GNT1-7 s’əʋa məχ’əmət’i ʔaχ’ɨħa tazəgad͡ʒɨw s’əba məχ’əmət’i ʔaχ’ɨħa t-a-zəgad͡ʒɨw milk VN-seat object IPV-CAUS:2SM-prepare ‘You prepare an object to put in milk.’ DMA6-13 ʔɨh ʔɨh PART ‘I am attending you attentively.’ ASGEDE 98 GNT1-8 bɨmaj bɨdəmbi gəʋɨrχa tɨħas’ɨʋo bɨ-maj bɨdəmbi gəbɨr-χa tɨ-ħas’ɨb-o INST-water very well do:CN-SJ:2SM IPV:SJ:2SM-wash:IMP-OBJ:3SM ‘You wash it very well with water.’ DMA6-14 ʔɨ…ʃʃi ʔɨ…ʃʃi Okay ‘That is interesting; would you please proceed to describe.’ GNT1-9 kaʔu [tɨʕat’ɨno] {ʔɨ…} #bɨʔawlɨʕ tɨʕat’ɨno# kaʔu tɨ-ʕat’ɨn-o ʔɨ… bɨ-ʔawlɨʕ tɨ-ʕat’ɨn-o then SJ:2SM-steam:IPV-OBJ:3SM RPR INST-olive SJ:2SM-steam:IPV-OBJ:3SM ‘Then, you steam it with an olive.’ The repair in example (16) above aims at inserting a missing constituent. It is marked by a filler ʔɨ… followed by about two seconds’ pause. The process, however, does not just repair but has also a planning process. The speaker extends the pause because she wanted to buy time to recall the missed linguistic constituent bɨʔawlɨʕ ‘by olive (tree)’. 4.2.3 Particles A particle refers to a linguistic unit comprising small words that in most cases are uninflected (Asgede, 2019). The term is used as a collective term for the linguistic expressions like dəʔa ‘rather’, and ʔɨbba ‘(I) mean’. The particle ʔɨbba ‘I mean’, which has different functions in different contexts, initiates repair in the following example (17)ː (17) SPNT1-64 [ħagos] {ʔɨbba} #ʃəgunu# məs’ɨʔudə ħagos ʔɨbba ʃəggunu məs’ɨʔ-u-də NAME RPR NAME come:PRV-3SM ‘Did Hagos I mean Shegunu come?’ In example (17), the particle ʔɨbba ‘I mean’ appeared between two proper nouns (Hagos and Shegunu). This particular particle functions as a replacive marker from the possible alternative names that the speaker knows and associates with another. Such wrong constituents and their corrected constituents appear as part of the conversation disfluencies because the constituents have some close associations in the speaker’s mind. Here in the above example, the two proper nouns name two brothers; thus, the speaker uttered the first one wrongly, which is repaired later. The repair initiation that is ʔɨbba ‘I mean’ is a particle. Another particle that could have replaced the mentioned particle is dəʔa ‘rather’ which is also realized as dəʔam ‘rather’ in the corpus. The latter is used to initiate as a speaker repairs a portion of a prior utterance that asks for confirmation by oneself. For example, ħagos dəjɨʔu dɨməs’ɨlo ʃəgunu? ʃəgunu dəʔa ‘Is that Hagos or Shegunu who is coming? It is rather Shegunu’ is a typical example that depicts how dəʔa ‘rather’ is used to initiate a repair. The difference between these two particles is that the first one is used before the repaired part of an utterance but the latter one appears after the repaired part of an utterance. These instance functions of the two particles ensure that particles are used to initiate repair by Tigrinya speakers. 4.2.4 Predicates Predicates in this paper are consumed as words and/or phrases that have the qualities that a verb has in Tigrinya. For example, the verbal noun malətəj ‘I mean’, the verbal phrases ʔajχonχuj ‘I am not’, ʔɨjʔoj ‘No’ are examples of linguistic constituents consumed as predicates. A speaker may also reformulate part of her/his speech after uttering a whole word. Unlike cut-offs, lexical constituents index repair, which implies that a speaker gets cognizant that she or he needs to mark a reformulation after a word is uttered. SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 99 (18) DM1-20 gɨdəj hɨndej ʕamətu jɨʔu bɨsruχə məʔazi jɨʔu dɨtwələdə gɨdəj hɨndej ʕamət-u jɨʔ-u bɨ-sr-u-χə məʔazi jɨʔ-u NAME how many year:POSS:3SM COP-3SM ABL-root-3SM-Q when COP-3SM dɨ-t-wələd-ə REL-PRV-birth:SJ:3SM ‘How old is Gidey? When was he born?’ MA1-61 [bɨzəmənə ʕɨnt͡ ʃ’ɨwa ʔɨkko jɨʔu dɨtwələdə] bɨ-zəmən-ə ʕɨnt͡ ʃ’ɨwa ʔɨkko jɨʔ-u dɨ-t-wələd-ə by-time-of rat FOC COP-3SM REL-PRV-birth:SJ:3SM ‘It is on the rat’s era.’ MA1-62 {ʔɨjʔoj} #zəmənə ʕɨnt͡ ʃ’ɨwa ħalɨfus bɨχɨltə ʕamətu dəʔa məsələni ʔatta# ʔɨjʔoj zəmən-ə ʕɨnt͡ ʃ’ɨwa ħalɨfus bɨ-χɨltə ʕamətu dəʔa no time-of rat pass:PRV-3SM-FOC by-two year-POSS:3SM FOC məsəl-ə-ni ʔatt-a look:IPV-OB:3SM-SJ:1S VOC-2SM ‘No, I think it is actually two years after the Rat’s era.’ The constituent ʔɨjʔoj ‘no’ initiates a repair. Here the repairable is one complete sentence that is segmented as one discourse unit. It indicates that the previous segment is wrong and should be repaired. It indexes that the host segment is a repaired form of the trouble source. The linguistic constituent ʔɨjʔoj ‘no’ points back to the original segment and fore to the altered segment. It, therefore, indexes the host segment is repaired. Repair is also made even after long trouble source constituents of a discourse are articulated as can be seen in example (18) above. Based on the mini-corpus, ʔajχonχuj ‘No! I am not’ also has a similar function with ʔɨjʔoj ‘no’. This linguistic unit (ʔajχonχuj ‘No! I am not’) has a morpheme that marks an agreement (ʔaj-χon-χu-j NEG-become:IPV-1S-NEG). This phrasal constituent initiates that the prior segment was inappropriate and a revised replacement for it is to come. 4.2.5 Pauses and Non-Verbal Signals Long pauses may also indicate that a speaker is about to repair a portion of an original problematic utterance. Furthermore, nonverbal signals such as facial expressions and gestures, which usually accompany verbal repair initiators, can independently indicate that a repair is about to take place. Nodding left and right, for example, may indicate that the speaker recognizes his original speech is problematic and intends to replace it. Furthermore, twinkling one’s eyes or closing both eyes may indicate that the speaker is about to repair a part of a problematic original utterance. 4.3 Results of Self-Repair As highlighted in the introduction, self-repair refers to self-correcting an error or inappropriate expression of an original part of an utterance. This conversational feature allows to expand an idea, replacing a constituent or an entire turn, reorder constituents, restart a turn or part of a turn, and insert a forgotten but relevant constituent. Repair results Total Number of Cases Identified Percentage Expanding 92 23 Replacing 91 22.75 Re-ordering 82 20.5 Restarting 70 17.5 Inserting 65 16.25 Total 400 100% Table 5. List of self-repair results. ASGEDE 100 E x p an d in g R ep la ci n g R e- o rd er in g R es ta rt in g In se rt in g S u m Sum 328 363 331 335 243 1600 P 0.205 0.227 0.207 0.209 0.152 Pe 0.203 Po 0.733 K 0.665 Table 6. Inter-rater agreement on the results of self-repairs. As depicted in Table 6, the inter-rater agreement is 0.665 which is interpreted as substantially relevant agreement according to Fleiss’ kappa (see Table 10 in Appendix III). After conducting a detailed investigation, the author found that expanding (92 cases) and replacing (91 cases) are the most frequently observed results of self-repair in Tigrinya. The description of each self-repair result will be presented in the following pages. 4.3.1 Expanding By expanding, a speaker seeks to clarify or reformulate how a concept was communicated in a previous turn. Such justifications or rephrasing are added by speakers to facilitate smooth contact with their correspondents. For instance, an explanation might be significant to clarify concepts, and rephrasing might be relevant to present ideas in a new way so that the listener has other options for comprehending the argument made by the utterances. malət ‘means’, is a very frequently used lexeme in Tigrinya to convey a further elaboration. In real speech, malət, ‘means’, is most often realized as malətəj ‘I mean’. (19) BLY1-21 ʃɨʋa dɨʔɑχonuj ʃɨba dɨʔɑ-χon-u-j paralize FOC-become:IPV-SJ:3PLM-FOC ‘They rather are unskilled (lit.They rather are paralyzed.). BLY1-22 malətəjssi suχ’lom mɨt’rom ʔaħɨʋt’om haməj wulom kɨsərɨħu mɨwdɨldal dəʔam malət-əj-ssi suχ’-bɨl-om mɨt’ri-om ʔa-ħɨbt’-om means-POSS:1S-FOC quiet-say:IPV-SJ:3PLM buttocks CAUS-fat:PRV-SJ:3PLM haməj wul-om kɨ-sərɨħ-u mɨ-wdɨldal dəʔam How say:CN-3PLM IPV:3-work-SJ:3PLM VN-idle FOC ‘I mean that they can’t do a job except wondering here and there with their big physical appearance.’ The lexeme malətəj ‘I mean’ in example (19) above indicates to the listener that the speaker is about to reformulate the way an idea was described. An explanation and justification of the prior discourse unit’s claim is presented in the turn that hosts the reformulation initiator lexeme. The ‘students’ that are the issue of the discussion, are ineffective in assisting their parents because the students lack the skill and gut and that is described as ‘they are paralyzed’. The speaker goes on to explain that the students are weak and prefer to wander here and there rather than be engaged in some livelihood activities on which their parents rely. In the repairing segment, the speaker explains why he described them as “paralyzed” in the repairable segment. He means that, in contrast to their chiseled bodies, they are unqualified and uninterested in assisting their parents in any way. The verbal noun malətəj ‘I mean’ appears at the initial position of the repaired segment as depicted in example (19) above. It also appears to be a final position of the repaired constituent but before a copular verb as in suχ’lom mɨt’rom ʔaħɨʋt’om haməj wulom kɨsərɨħu mɨwdɨldal SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 101 dəʔam malətəj jɨʔu ‘I mean that they can’t do a job except wondering here and there with their big physical appearance.’ 4.3.2 Replacement The second type of repair result is a replacement which refers to replacing a wrongly uttered constituent with a correct one. This is initiated by different linguistic units. Predicative lexemes such as ʔajχonχuj ‘no, I am not’ and particles such as ʔɨbba ‘I mean’ are the most common lexical devices that initiate a replacement. (20) DM1-96 ʔajjajjə ʔajəʕawuji məʔazi jɨʔu dɨmotə ʔajja-jjə ʔajə-ʕawuji məʔazi jɨ-ʔu dɨ-mot-ə father-RL:1S father-big when COP-3SM REL-death:PRV-SJ:3SM ‘Father, when did Grandpa die?’ MA1-221 ʕasərtə ʕamətu χojnuwwo ʔasərtə ʕamət-u χojn-u-ww-o ten year-poss.3sm become:IPV-SJ.3SM-OBJ.3SM ‘Ten years have passed.’ (lit. It has been ten years.’ MA1-222 ʔajχonχuj ʕasərtə ħamuʃtə ʕamətu dəʔatta ʔaj-χon-χu-j ʕasərtə ħamuʃtə ʕamət-u dəʔa-tta NEG-become:IPV-1S-NEG Ten Five year-POSS.3SM DM_FOC-VOC.2SM ‘It was rather fifteen years ago.’ As can be seen in example (20) above, ʔajχonχuj ‘No, I am not’ indicates that the following utterance is going to replace a prior portion of discourse. So, it is not 10 years since the grandpa is died but 15 years. (21) SPNT1-127 ħagaj ħagaj ʔɨbba χɨrəmti χɨrəmti ʔabəj laħləfχajjo jɨʔu hɨndizi dɨt’əfaʔχa ħagaj ħagaj ʔɨbba χɨrəmti χɨrəmti ʔabəj winter winter RPR summer summer where la-ħləf-χa-jj-o jɨʔ-u hɨndi-zi dɨ-t’əfaʔ-χa PROG-spend:IPV-SJ.2SM-Ø-OBJ.3SM COP-3SM much-PRX.3SM REL-lost:PRV-SJ.2SM ‘Where have you been spending every winter I mean every summer that you have been missing for so long? .’ The particle ʔɨbba ‘rather, I mean’ in this particular context is another replacive marker. It appears between the problematic portion of the prior discourse unit and its corrected constituents. As can be seen in example (21) above, ʔɨbba ‘rather, I mean’ appeared between ħagaj ħagaj ‘every winter’ and χɨrəmti χɨrəmti ‘every summer’ so that the latter one is the constituent that replaces the former ħagaj ħagaj ‘every winter’, that is, the repairable one. 4.3.3 Insertion Insertion refers to adding a constituent which was missed in the repairable portion of an utterance. The inserted constituent can be a word, phrase, clause and/ or sentence. It may be initiated by cut-offs, fillers, particles, or predicative words. (22) GNT1-10 kaʔu laχadɨnχa guħatɨj mɨʃətɨj s’əʋa tawahlɨləllu kaʔu la-χadɨn-χa guħat-ɨj mɨʃət-ɨj s’əba then PROG-close:CN-3SM morning-and evening-and milk ta-wahlɨl-ə-ll-u cAUS-store-SJ.2SM-BEN-OBJ.3SM ‘Every morning and evening, you store (milk) in it.’ ASGEDE 102 GNT1-11 bɨdawahləlɨχa tɨħak’uno s’ɨnaħ medʒəmerɨja tarɨgʔo dəʔa kaʔu tɨħak’uno bɨd-a-wahləl-ɨχa tɨ-ħak’un-o s’ɨnaħ medʒəmerɨja after-CAUS-store-SJ.3SM IPV-churn:SJ.2SM-OBJ.3SM wait:SJ.2SM first ta-rɨgʔ-o dəʔa kaʔu tɨ-ħak’un-o IPV-CAUS-coagulate:SJ.2SM-OBJ.3SM rather then IPV-churn:SJ.2SM-OBJ:3SM ‘After you store it, you churn it; wait, first you let it coagulate rather then churn it.’ The particle dəʔa ‘rather’ that functions, in most cases, to selectively emphasize a constituent, appears next to a linguistic constituent that replaces a problematic portion of a prior discourse. In the above excerpt, the speaker uttered s’ɨnaħ ‘wait’ which directly tells us something went wrong with the prior discourse; it, however, doesn’t directly index a repair. Next to that, the speaker added a new step medʒəmerɨja tarɨgʔo ‘first you let it coagulate’ that should precede a step that was stated in the prior constituent of the turn tɨħak’uno ‘you churn it’. As you can see in the excerpt, the repaired segment is followed by the selective repair indicator. Syntactically, this particle is different from the particle ʔɨbba ‘rather, I mean’. The former appears next to the repairing segment whereas the latter appears between the repairable and the repairing segment. 4.3.4 Restart Restart, at least in this paper, refers to a conversational strategy by which a speaker stops/aborts speaking and then restarts either with a modified version of the prior portion of discourse or with a new idea that is different from the trouble source. (23) BLY1-001 suməj gɨrmaj ħagos ʔə… ʔɨʃʃi suməj gɨrmaj ħagos ħagos ħuluf ħuluf ʔaləm ʔaləm s’adɨχ’ jɨʔu sum-əj gɨrmaj ħagos ʔə… ʔɨʃʃi suməj gɨrmaj ħagos name-POSS:1S NAME NAME RPR Okay name-POSS:1S NAME NAME ħagos ħuluf ħuluf ʔaləm ʔaləm s’adɨχ’ jɨʔ-u NAME NAME NAME NAME NAME NAME COP-3SM ‘My name Girmay Hagos ʔə… okay my name is Girmay Hagos Huluf Alem Tsadik.’ As can be seen in example (23) above, the speaker aborts speaking about himself and restarts after he has taken his time to reorganize his speech. The speaker gets stuck into uttering his grandpa’s name and thus stops speaking. The speaker then restarts his talk with the same content as the trouble source segment. This was initiated by the filler ʔə which is followed by the stretched vowel sound that is marked by three dots (…). That again is followed by the particle ʔɨʃʃi ‘okay’ that can be considered as a false starter. The speaker reorganizes his idea and restarts. However, as we can see from example (24) below, the speaker aborts speaking in a plan to come up with a different point of discussion. The speaker stops talking about the previous topic and shifts the point of discussion to another. Such a repair is indicated by ħɨdəgo ʔɨsti ‘please leave it’ which states that the speaker is no more interested in talking about the issue. That again closed with the enclitic =wa that is attached to the last constituent of the repairable segment. The purpose of such repair to restart is to change a point of discussion – here from talking about the challenges the speaker is facing to asking about the health of the correspondent’s family. (24) BLY1-111 ʔɨttom ʔaħwatəj bɨgɨrat tɨs’alʔom ʔajnəgagəruj ʔɨtt-om ʔa-ħaw-at-əj bɨ-gɨrat tɨ-s’alʔ-om DST-3PLM PL-brother-PL-RL:1S INST-farmland PRV-oppose-SJ:3PLM ʔaj-nəgagər-u-j NEG-speak:REF-3PLM-NEG ‘My brothers do not speak to each other because they are in disagreement over a farmland.’ SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 103 BLY1-112 ʔaddəjjəllə mɨs ʔakkojjə bɨnaddɨʔom wursi tɨs’alʔom fɨrdi ʋət məlaləsləwu ʔaddo-jj-ə-llə mɨs ʔakko-jjə bɨ-na-addo-ʔ-om wursi mother-Ø-RL:1S-and with uncle-RL:1S INST-POSS-mother-Ø-3PLM heirloom tɨ-s’alʔi-om fɨrdi bət məlaləs-ʔɨllə-w-u REC-quarrel-3PLM Justice house ambulation-existːIPV-SJ:3PLM ‘My mother is currently in a quarrel with my uncle, and they are going to court.’ BLY1-113 ħɨdəgo ʔɨsti ʔanə ʔagnɨjəjjo dɨnhəχu gudwa… ʕɨjjal haməj ʔɨnhəwu ħɨdəg-o ʔɨsti ʔanə ʔa-gnɨj-ə-jj-o leave.IMP:SJ.2SM-OBJ.3SM part 1S CAUS-find:CN-SBJ.1S-OBJ.3SM dɨ-ʔɨnh-ə-χu gud=wa… REL-exist:PRV-OBJ.3SM-SJ.1S bad thing-RPR ‘Please leave the people I am dealing with…, how are all your family members?’ Besides, a speaker repairs a portion of a prior segment to restart an utterance to shift from sticking to one detail to another without leaving the major point of discussion. The last turn of example (25) below has a particle wa which can be roughly translated as ‘whatever’, and is used to initiate a shift from talking about the weakness of the students to a possible alternative that they may excel at, that is, being clever students. (25) BLY1-21 ʃɨba dɨʔɑχonuj ʃɨba dɨʔɑ-χon-u-j paralize FOC-become:IPV-SJ:3PLM-FOC ‘They are rather unskilled’ (lit.They are are paralyzed.). BLY1-22 malətəjssi suχ’lom mɨt’rom ʔaħɨʋt’om haməj wulom kɨsərɨħu mɨwdɨldal dəʔam malət-əj-ssi suχ’-bɨl-om mɨt’ri-om ʔa-ħɨbt’-om means-POSS:1S-FOC quiet-say:IPV-SJ:3PLM Buttocks CAUS-fat:PRV-SJ:3PLM haməj wul-om kɨ-sərɨħ-u mɨ-wdɨldal dəʔa-m how say:CN-3PLM IPV:3-work-SJ:3PLM VN-idle FOC-FOC ‘This (pointing at a farming field) requires three pairs of oxen to plow.’ BLY1-23 ʔazatomu haməj… wa… ʔatta tɨmhɨrtom hamgobəzullə t͡ ʃ’əllə neʋɨru ʔaz-atomu haməj… wa… ʔatt-a tɨmhɨrti-om PRX.3PLM how RPR VOC-2SM education-POSS:3PLM ham-gobəz-u-llə t͡ ʃ’əllə nebɨr-u if-excel:IPV-SJ.3PLM-and Okay exist:PRV-3SM ‘How could they …whatever… it could have been fine if they had excelled in their education.’ In example (26) below, the speaker restarts the turn at the fifth constituent of the turn. The repair operation results in deleting or leaving out some non-relevant portion of the trouble source. Thus, the first four constituent words of the trouble source, i.e., ʔɨndɨr haftam tɨχon dalχamma ‘see, if you want to be rich’ are replaced by haftam dɨmuχʷan ‘to be rich’ by a restarting mechanism; while two constituents are deleted, one is modified and another constituent is maintained as it originally was in the repairing segment. (26) MA2-78 ʔɨndɨħɨr haftam tɨχon dalχamma haftam dɨmuχʷan hɨras dəjmɨbzaħ jɨʔu ʔɨndɨħɨr haftam tɨχon dalɨj-χa-mma haftam dɨ-muχʷan If rich IPV:2-become want:IPV-SJ:2SM-FOC rich PUR-become hɨras dəjmɨbzaħ jɨʔ-u sleep NEG-multiple COP-3SM ‘If you really want to be rich, to be rich, you can not sleep too much.’ 4.3.5 Reordering Constituents might appear in a portion of a prior utterance as a source of trouble that results in reordering them. As can be seen in example (10), a sentence or long constituents may take the wrong order in speech. ASGEDE 104 The most common wrong order found in the data is short constituents like words as can be seen in example (27) below. (27) MA1-143 ʔɨsumma bəlaj ʔadanə ʔɨbba ʔadanə bəlaj jɨʔu sɨmu ʔɨsu-mma bəlaj ʔadanə ʔɨbba ʔadanə bəlaj jɨʔ-u sɨm-u 2SM-FOC NAME NAME mean NAME NAME COP-3SM name-POSS:3SM ‘That one’s name is BelayAdane I mean Adene Belay.’ Here in this example, the trouble source is the wrong order of the first name and middle. The self-repair initiated by the particle ʔɨbba ‘(I) mean’ results in reordering the wrongly ordered nouns in the repairable portion of the utterance. 5 Discussion and Recommendation Like many language speakers in the world (Fox et al., 2013), Tigrinya speakers often repair portions of their prior utterances that appear to be ‘inappropriate’. The inappropriateness of a portion of an utterance may or not have an association with ‘error’. Through data analysis, 400 repairs have been identified. These identified repairs resulted from various sources of trouble. The almost perfect Fleiss’ kappa inter-rater agreement (0.835) indicates that the most common sources of repair are pronunciation (0.218), diction (0.173), missing constituents (0.159), wrong constituents (0.125), word order issues (0.124), misunderstandings (0.106), and focus misplacement (0.088). It was observed that the pronunciation and diction issues are related with dialect differences. The paper highlights that more than half of the diction problem sources for self-repair is not related to dialectical variation. These sources of repair, however, need a further study on statistically showing to what degree dialect variation impacts pronunciation and diction issues in speech. Pronunciation-related problems resulted from tongue slip and the dialect variation. The data was recorded in the presence of the researcher, who had been away from the area for about 15 years. It seems that the participants (wrongly) perceived that the researcher preferred to hear the lexemes as they are pronounced in the quasi-standard Tigrinya used in education, media, and offices at various levels; six of the recorded participants explained this while we held informal discussions about what they repair and replace words for. A participant might repair lajəs’a to nab jəs’a ‘to Yetsa’ and tɨχəd to tɨχəjɨd ‘you go’. How such issues work in the conversations of speakers of other varieties of the language (and other languages spoken in Ethiopia too) is left for further research. In addition to the dialectical variation, triggered repair to replace a lexical item, interpersonal relationships among participants is another trouble source. For example, the constituent ħɨləf ‘(you:2SM) pass through’ is replaced by ħɨləfu ‘(you:2SM.H) pass through’ for the participants’ social status variation; tɨħarɨʔ is replaced by tɨftɨn ‘urinate’ for diction purposes that have implications on the participants’ dialect variation background. When speakers speak out constituents such as improper nouns in a specific context, they use the particle ʔɨbba between the repairable and the repaired which indexes repair. This uncovers the fact that Tigrinya honorific grammatical feature has impact on how interlocutors interact with one another. Different linguistic devices initiate self-repair. Cut-offs that do not directly index self-repair as well as filled pauses which have multiple functions in conversation are used. Cut-offs that are the dominant self- repairs (0.593) have other functions: planning process being the most common one (Asgede, 2019). There is a need to study how cut-offs are distributed and used in speech independently. Predicative phrases like ʔajχonχuj ‘I am not’ and ʔɨjʔoj ‘no’ as well as particles such as ʔɨbba ‘I mean’ and dəʔa ‘rather’ are used to initiate self-repair. As stated in different empirical studies in different languages (Ginzburg et al., 2007), self-repair also is done without any repair marker; initiators are therefore not mandatory in Tigrinya too. That seems the reason why pauses and nonverbal cues that speakers may leave out of their utterance can initiate self-repairs. These findings match with what Shriberg (1994) found out, that lexical items and disfluencies such as filled pauses initiate self-repair. This also implies that self-repair can be ‘explicitly’ marked by linguistic and non-linguistic constituents as mentioned by Meteer et al. (1995). SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 105 Different types of self-repair result in different operations. For example, self-repair may be used to expand an idea presented in a previous section of discourse, which is most commonly initiated by the verbal noun malət ‘means’. Self-repair may also result in replacing a wrongly uttered constituent, initiated by predicative phrases such as ʔajχonχuj ‘I am not’ and ʔɨjʔoj ‘no’ and the particle ʔɨbba ‘I mean’. A particle dəʔa ‘rather’ initiates the insertion of a missing constituent. Finally, though it would have been possible to study the phonetics features of every repair initiator and its categorizations, how self-repair relates with the syntactic features of Tigrinya, repair in the process of language acquisition and learning, thorough discussion on the relationship between self-repair and the demographic features of participants, etc., they are left aside and the author recommends such issues for further investigation. References Dagnew Mache Asgede (2007). Tigrinya Dialectal Variations between Rayya and Adwa. Arba-Minch University: Senior Essay. Dagnew Mache Asgede (2019). Discourse Markers in Rayya Tigrinya: Documentation and Linguistic Analysis. Dissertation. Addis Ababa University. http://etd.aau.edu.et/handle/123456789/24899 Dagnew Mache Asgede (2023). Backchannels in Tigrinya. Ethiopian Journal of Business and Social Science, 6(1), pp. 15-35. https://doi.org/10.59122/144F59lk David M. Bell (2010). Nevertheless, Still, and Yet: Concessive Cancellative Discourse Markers. Journal of Pragmatics, 42, 1912-1927. Emily M. Bender and Batya Friedman (2018). Data Statements for Natural Language Processing: Toward Mitigating System Bias and Enabling Better Science. Transactions of the Association for Computational Linguistics, 6:587–604. Elisabeth R. Blackmer and Janet L. Mitton (1991). Theories of monitoring and timing of repairs in spontaneous speech. Cognition, 39:173–194. Hatte Anne Blejer (1986). Discourse Markers in Early Semitic, and Their Reanalysis in Subsequent Dialects. Dissertation. The University of Texas at Austin. Susan Brennan and Maurice Williams (1995). The Feeling of another’s Knowing: Prosody and Filled Pauses as Cues to Listeners about the Metacognitive States of Speakers. Journal of Memory and Language, 34:383–398. Susan Brennan and Michael Schober (2001). How listeners compensate for disfluencies in spontaneous speech. Journal of Memory and Language, 44:274–296. James Deese (1984). Thoughts into Speech: the Psychology of a Language. Prentice-Hall, Englewood Cliffs, NJ. Vincent A. de Rooji (2000). French Discourse Markers in Shaba Swahili Conversations. The International Journal of Bilingualism, 4.4:447-467. Joseph L. Fleiss (1971). Measuring nominal scale agreement among many raters. Psychological Bulletin, 76:378-382. http://etd.aau.edu.et/handle/123456789/24899 https://doi.org/10.59122/144F59lk ASGEDE 106 Joseph L. Fleiss, Bruce Levin, and Myunghee C. Paik (2003). Statistical methods for rates and proportions (3rd ed.). Hoboken, NJ: Wiley. Barbara A. Fox, Makoto Hayashi, and Robert Jasperson (1996). Resources and repair: a cross-linguistic study of syntax and repair. In E. Ochs, E. A. Schegloff, & S. A. Thompson (Eds.), Interaction and Grammar (pp. 185–237) chapter, Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511620874.004 Barbara A. Fox, Trevor Benjamin, and Harrie Mazeland (2013). Conversational Analysis and Repair Organization: Overview. In C. Chapelle (Ed.), The Encyclopedia of Applied Linguistics. https://onlinelibrary.wiley.com/doi/10.1002/9781405198431.wbeal1314 Jonathan Ginzburg, Raquel Fernández, and David Schlangen (2007). Unifying self-and other-repair. In Decalog 2007: Proceedings of the eleventh Workshop on the Semantics and Pragmatics of Dialogue, May 30-June 1 2007, Rovereto, Italy. Abraham Girmay (2012). Language Shift in Progress in the Raya Variety of Tigrinya: The Case of Alamata and Kobo Weredas. MA Thesis. Addis Ababa: Addis Ababa University. Kieu-Phuong Ha and Martine Grice (2017). Tone and Intonation in Discourse Management-How Do Speakers of Standard Vietnamese Initiate a Repair? Journal of Pragmatics, 107:60-83. Patrick G. Healey, Marcus Colman, and Mike Thirlwell (2005). Analysing Multimodal Communication: Repair-Based Measures of Human Communicative Coordination. Advances in Natural Multimodal Dialogue Systems, 113-129. Peter A. Heeman and James F. Allen (1999). Speech Repairs, Intonational Phrases, and Discourse Markers: Modeling Speakers’ Utterance in Spoken Dialogue. Association for Computational Linguistics, 25.4, 528-571. Jim Hlavac (2011). Hesitation and Monitoring Phenomena in Bilingual Speech: A Consequence of Code- switching or a Strategy to Facilitate its Incorporation? Journal of Pragmatics, 43:3793–3806. https://doi.org/10.1016/j.pragma.2011.09.008 Julian Hough (2015). Modelling Incremental Self-Repair Processing in Dialogue. Unpublished PhD thesis. Queen Mary University of London. http://qmro.qmul.ac.uk/xmlui/handle/123456789/9094 Farhat Jabeen, Asim M. Asim Rai, and Sara Arif (2011). A Corpus-Based Study of Discourse Markers in British and Pakistani Speech. In International Journal of Language Studies (IJLS), 5(4):69-86. Masahito Kawamori, Takeshi Kawabata, and Akira Shimazu (1998). Discourse Markers in Spontaneous Dialogue: A Corpus-Based Study of Japanese and English. In Manfred Stede, Leo Wanner and Eduard Hovy (eds.), Discourse Relations and Discourse Markers: Proceedings of the Workshop, 93-99. Canada: Montreal: Universite de Montreal. Tsehaye Kiros (2009). A Comparison of Wajerat Tigrinya Vs. Standard Tigrinya. MA Thesis: Addis Ababa: Addis Ababa University. Willem J. Levelt (1983). Monitoring and self-repair in speech. Cognition, 14(4):41–104. https://doi.org/10.1017/CBO9780511620874.004 https://onlinelibrary.wiley.com/doi/10.1002/9781405198431.wbeal1314 https://doi.org/10.1016/j.pragma.2011.09.008 http://qmro.qmul.ac.uk/xmlui/handle/123456789/9094 SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 107 Yaron Matras (2000). Fusion and the Cognitive Basis for the Bilingual Discourse Markers. The International Journal of Bilingualism, 4.4, 505-528. Douglas W. Maynard (2013). Everyone and No One to Turn to: Intellectual Roots and Contexts for Conversation Analysis. In Jack Sidnell and Tanya Stivers (eds.), The Handbook of Conversation Analysis, 11-31. West Sussex: Wiley-Blackwell. Niguss Weldezgu Mehari (2011). The Impacts of Rayya Dialectal Variations and the Influence of Amharic on Medium of Instruction: The Case of Alamata Primary Schools. Addis Ababa University.MA Thesis. http://etd.aau.edu.et/handle/123456789/16280 Niguss Weldezgu Mehari. (2021). A Grammar of Rayya Tigrinya. Addis Ababa University. Dissertation. Marie W. Meteer, Ann A. Taylor, R. MacIntyre, and R. Iyer (1995). Dysfluency annotation stylebook for the Switchboard corpus. Philadelphia, PA: University of Pennsylvania. Lorenza Mondanda (2013). The Conversation Analytic Approach to Data Collection. In Jack Sidnell and Tanya Stivers (eds.), The Handbook of Conversation Analysis, 32-56. West Sussex: Wiley-Blackwell. Andrew Murphy (2019). Resolving conflicts with violable constraints: On the cross-modular parallelism of repairs. Glossa: a journal of general linguistics, 4(1): 9. 1–39. https://doi.org/10.5334/gjgl.608 Christine H. Nakatani and Julia Hirschberg (1994). A corpus-based study of repair cues in spontaneous. speech. Journal of the Acoustical Society of America, 95(3), pp. 1603-1616. https://doi.org/10.1121/1.408547 Gessesse Nigusse (2012). Language Use in the Criminal Justice Process: The Case of Raya Alamata Woreda in Tigray Region. MA Thesis. Addis Ababa: Addis Ababa University. Emanuel A. Schegloff (1992). Repair after next turn: The last structurally provided defense of intersubjectivity in conversation. American Journal of Sociology, V-97: pp. 1295–1345. https://www.journals.uchicago.edu/doi/10.1086/229903 Emanuel Schegloff, Gail Jefferson, and Harvey Sacks (1977). The preference for Self-correction of Repair in Conversation. Language, 53(2):361-382. https://doi.org/10.2307/413107 Lawrence Schourup (1985). Common Discourse Particles in English Conversation: ‘like’, ‘well’, ‘y’know’. New York: Garland. Elizabeth E. Shriberg (1994). Preliminaries to a theory of speech disfluencies. Ph.D. thesis, University of California at Berkeley, Berkeley, USA. Tanya Stivers and Jack Sidnell (2013). Introduction. In Jack Sidnell and Tanya Stivers (eds.), The Handbook of Conversation Analysis, 1-8. West Sussex: Wiley-Blackwell. Marc Swerts (1998). Filled pauses as markers of discourse structure. Journal of Pragmatics, 30(4):485- 496. https://doi.org/10.1016/S0378-2166(98)00014-9 Tsehaye Teferra (1979). Reference Grammar of Tigrinya. Doctoral Dissertation. Washington, D.C.: Georgetown University. http://etd.aau.edu.et/handle/123456789/16280 https://doi.org/10.5334/gjgl.608 https://psycnet.apa.org/doi/10.1121/1.408547 https://www.journals.uchicago.edu/doi/10.1086/229903 https://doi.org/10.2307/413107 https://psycnet.apa.org/doi/10.1016/S0378-2166(98)00014-9 ASGEDE 108 Yan Wang (2011). A Discourse-Pragmatic Functional Study of the Discourse markers Japanese Ano and Chinese Nage. Intercultural Communication Studies, 20.2, 41-61. Nigel Ward (1998). Some Exotic Discourse Markers of Spoken Dialog. In Manfred Stede, Leo Wanner and Eduard Hovy (eds.), Discourse Relations and Discourse Markers: Proceedings of the Workshop, 62- 64. Canada: Montreal: Universite de Montreal. Jean Wong and Hansun Z. Waring (2010). Conversation Analysis and Second Language Pedagogy: A Practical Guide to ESL/EFL Teachers. Routledge. New York and London. Robin Wooffitt (2005). Conversation Analysis and Discourse Analysis: A Comparative & Critical Introduction. SAGE Publications. London, Thousand Oaks & New Delhi. Erkan Yilmaz (2004). A Pragmatic Analysis of Turkish Discourse Particles: Yani, İşte and Şey. Doctoral Dissertation: Middle East Technical University. Tesfay Tewolde Yohannes (2002). A Modern Grammar of Tigrinya. Rome, Tipografia U.Detti-Via G.Savonarola. Appendix I: Transcription Conventions and Symbols The data was transcribed and annotated phonetically; the detailed annotation that includes gloss for every morph and free translation is helpful for the readers who do not speak and read Tigrinya. For readability, the transcription conventions used in the excerpts presented in this paper are the following: 1 first person 2 second person 3 third person ALL allative BEN beneficiary CAUS causative CN converb CNVRPR conversational process COP copular DEF definiteness DEST distal DM discourse marker F feminine FOC focus H honorific IMP imperative INST instrument IPV imperfective JUSS jussive lit. literal meaning LOC locative M masculine NEG negative Ø zero meaning OBJ object PART particle PL plural PLNPRC planning process PLT polite POSS possessive PROG progressive PRX proximal PUR purpose PVR perfective Q question marker REF reflexive REL relativized RL relative (affection) RPR repair S singular SJ subject VN verbal noun VOC vocative NAME a gloss for proper nouns / overlap begins .. cut-off = clitic {} repair mechanism - morph and gloss boundary SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 109 : boundary for grammatical functions by one morph … long pause [] trouble source # # repair solution \ another function Appendix II: Data Statement During a three-month fieldwork during which the researcher recorded the speech data, the mini-corpus temporarily named as Rayya Tigrinya mini-speech corpus has been established. Originally, the mini-corpus totalled 24 hours of recorded audio/video data. For research ethics purposes, based on the basic schema recommended by Bender and Friedman (2018), the data statement is detailed as follows. CURATION RATIONALE: With the help of three research assistants (two with a BA degree in English Language and Literature and one with a BA degree in Tigrinya) and the consent of the participants, the researcher selected six hours of oral data. Then the selected six hours of oral data was segmented and transcribed by the three individuals mentioned above and by the researcher. Later on, the researcher edited and made some corrections on the transcription and made sample tagging of repairs. The article is entirely based on audio and video-recorded oral data. The data is made up of casual conversations, informal interviews, and stories. Spontaneous conversations, informal interviews that were recorded on a variety of topics including how to do things, biographies, tales, explanations of some cultural performances such as shadey, shiwulalo, gebiche, and gebeta, farming, family relationships, and how to culturally mediate conflicts were selected; these accounts lasted approximately one hour and forty-two minutes. Ten spontaneous conversations that were recorded at a market, a conflict resolution site, a bus station, a workplace, and a hotel are selected: the longest interactional conversation was twenty minutes long, while the shortest was four minutes and twenty-two seconds long. Types of oral data Number of recordings Length in minutes Total number of words Percent Number of repairing Repairing in Percentage Remark Spontaneous conversation 10 61 12,600 17.074 156 39 2 up to 4 participants Conflict resolution 2 62 14,495 16.931 132 33 6 & 9 participants Informal interview 20 59 12,497 16.935 51 12.75 2 participants each How to do things 8 58 11,650 15.786 30 7.5 2 participants each Biography 12 60 12,504 16.943 17 4.25 1 participant each Tales 18 62 12, 050 16.328 14 3.5 2 participants each Total 362 75,796 100 400 100% Table 7. Total number of recordings for each genre and number of self-repairs. According to the data, 400 repairs have been identified; repairing is very common in spontaneous conversations, followed by conflict resolution, which account for 156 and 132 instances, respectively. Only 14 repairs were identified in the narrating tales from the recorded genre of data sources. The above table shows the distributions of the repairs in detail. ASGEDE 110 LANGUAGE VARIETY: The linguistic data that this article relies on is collected from the residents of Neksege, a locality located to the west of Maichew, Southern Tigray Administrative Zone, and is used in the analysis. These people speak a Tigrinya variety that is not being represented enough in education, media, and offices (Asgede, 2007 and 2019; Mehari, 2011). SPEAKERS’ DEMOGRAPHIC FEATURES: Data was recorded from 38 participants. Table 8, summarizes the demographic features of the participants whose speech was recorded for the purpose of this research. All the speakers gave their consent before each recording started and ended. Number Remark Number Remark 1. Age 20-30 31-40 41-50 50-60 >60 2 5 6 10 15 4. Religion Orthodox Protestant Muslim 36 1 1 Originally from Qobo Originally from Maichew 2. Gender Female Male 14 24 5. Ethnicity Tigrean Agew Amhara 35 2 1 Originally from Ts’eta Originally from Qobo 3. Educational background Primary (0-4) Primary secondary (5-8) Highschool (9-12) College 15 9 8 6 6. Language background Native in Tigrinya Native in Amharic Native in Agew 36 1 1 8 of them speak Agew; other 5 speak English Speaks Tigrinya & English Speaks Tigrinya Total Number of Participants 38 Table 8. Demographic features of participants. ANNOTATORS’ DEMOGRAPHIC: As highlighted above, including the author of the current research article, the annotators are four in number. All the annotators are members of the speech community that the data is collected from. Four of them are native Tigrinya speakers. While the researcher speaks Amharic like a native, the others speak it with phonological, lexical and syntactic difficulties. They range in age from 35-48 and include one woman and three men. SPEECH SITUATIONS and SPEECH CHARACTERISTICS: As introduced in the previous pages, the data is recorded in different social settings. Except the data from conflict resolution, the remaining data was collected in informal settings and contains comics and several nonlinguistic features of daily interaction. The intended audience was therefore the participants who overtly showed up themselves in every dialogue and interview. The contents of the data are related to the living system of the community. RECORDING EQUIPMENT: The recording device used was an IC recorder, and some videos of conflict resolutions were video recorded using an Apple iPad. Appendix III: Inter-Annotator Agreement Based on an example text that contains 15 repairs, the researcher trained the annotators and raters for half a day. The content of the training was how to identify the repair typologies and how to annotate (transcribe) them. Based on Healey et al. (2005), a short guideline that encompasses the following points is given to the annotators. SELF-REPAIR IN TIGRINYA: TROUBLE SOURCES, MECHANISMS AND SOLUTIONS 111 A repair is identified as a self-repair, if the response to the following questions is ‘yes’ except for the fourth point:  Does the speaker of the current turn modify the original text one way or another by him/herself?  Is the speaker repairing his/her own original constituent/utterance?  Are you sure the current listener does not make any contribution to the request for a repair?  Is not the repair/revision completed by another person?  Does the speaker of the current turn edit, amend, or reprise the original text? To identify the trouble sources of self-repair, the raters should try to:  Identify why speakers repair the portion of their old speech. To identify the types of the repair mechanisms, consider the following points:  Note whether the repair marker (repair mechanism) is cut-off, particle, filled pause, noun, verb, pause, verb phrase, noun phrase, clause, facial expression, etc.  Note whether the purpose (result) of each identified self-repair is to expand, restart, insert, replace, reordering, etc. of the original constituent. To identify the result of repairs, the raters should try to:  Understand how the repair result is different from the repaired one,  Explain the contribution of the repair result on the discourse development,  Assign what the repair mechanisms entail on the repair result. Accordingly, the number of self-repairs identified from each genre by the annotators are shown in Table 9. This table only demonstrates the mean of the four annotators in comparison with the number of self-repair markers identified and used in this research. Genres Number of Self-repair Markers Identified by Annotators A1 A2 A3 A4 Mean Repair Markers Considered Spontaneous conversation 160 145 144 156 151 156 Conflict resolution 137 127 121 132 129 132 Interview 53 45 47 51 49 51 How to do things 34 30 27 30 30 30 Biography 18 16 15 17 17 17 Tales 18 12 10 14 14 14 Total 420 375 364 400 390 400 Table 9. Aannotators’ agreement on identifying repair markers. Considering the total number of repairs identified by the researcher and the remaining three annotators, the mean is 390. Therefore, the 400 repairs are accepted as valued and reliable. This is substantiated with the inter-rater agreement among the four raters (including the author) for the categories of self-repairs: trouble sources, repair mechanisms, and results of self-repairs under sections 4.1., 4.2., and 4.3., respectively. The Fleiss’ kappa model is employed to manually compute the inter-rater agreement for there are four raters (more than two raters) and the data is nominal (Fleiss 1971; 2003). 𝐾 = Po−Pe 1−Pe where K is Fleiss’ kappa; Po is observed agreement, Pe is expected agreement among the raters The inter-rater agreement for the sources of troubles in self-repair, repair mechanisms, and results of self- repairs is calculated in the analysis section using the Fleiss’ kappa interpretation intervals provided below in Table 10. ASGEDE 112 Fleiss’ kappa Interpretation <0.00 Poor agreement 0.00 to 0.20 Slight agreement 0.21 to 0.40 Fair agreement 0.41 to 0.60 Moderate agreement 0.61 to 0.80 Substantial agreement 0.81 to 1.00 Almost perfect Table 10. Interpretation of Fleiss’ kappa values.