The syntax of interjections in isiXhosa: A a corpus-driven study

This paper examines the syntactic properties of interjections in isiXhosa and their compliance with the interjectional prototype and its extra-systematicity as postulated in linguistic typology. By reviewing nearly two thousand uses of interjections in the comic genre, the authors conclude the following: in its integrity, the category of interjections is internally complex and diversified, containing members with varying degrees of canonicity and extra-systematicity. Although in various uses interjections comply with the interjectional prototype, being thus extra-systematic, in many others, their canonicity and extra-systematicity are significantly lower.


Introduction
Interjections constitute a universal word class (Ameka and Wilkins 2006: 7) and a typical feature of human language. Their relevance is particularly evident in isiXhosa (S 41) -a language that contains nearly three hundred and fifty fully grammaticalized and/or lexicalized interjectional items (Andrason and Dlali forthcoming). Although isiXhosa offers a formidable research arena, studies dedicated to interjections in this language are scarce and fragmentary. In most cases, researchers limit themselves to enumerating interjectional tokens (see Du Plessis 1978: 284-290) and presenting their meanings or functions (Jordan 1966: 204-205, Du Plessis 1978: 284-290, Oosthuisen 2016. 1 Sometimes, a few phonological (Oosthuisen 2016: 357) and morphological properties of interjections are identified (Oosthuisen 2016: 357-359, Kirsch, Skorge and Magona 1999: 236, Louw and Jubase 1963: 255-256, Schonstein Pinnock 1994. The area that is particularly under-researched in studies on isiXhosa interjections concerns syntax. All mentions of the syntactic properties of interjections can be reduced to a single statement: interjections "have no effect on [the] grammatical construction of a sentence" (Schonstein Pinnock 1994: 196) because they fail to enter into a relationship with, or exert effects on, its remaining components (Louw and Jubase 1963: 255;Oosthuisen 2016: 357). This statement never emerges from a process of rigorous analysis, nor is it ever substantiated by empirical evidence. 2 Although, as mentioned above, interjections constitute a pervasive linguistic feature, their visibility in language is uneven. Interjections are common in oral discourses (Romero-Trillo 2006: 640;Stange andNübling 2014: 1983;Stange 2016: 7). In contrast, their presence is much more limited in written discourse. Noticeable exceptions are the genres of written language that closely imitate spoken language -for instance, drama plays, emails, WhatsApp messages (Frehner 2008;Forster, Borgwaldt, and Neef 2012: 122;Stange andNübling 2014: 1983; Andrason and Dlali forthcoming) and comic strips (Fernández Cuesta 1990: 551;Nübling 2004: 16;de la Cruz Cabanillas and Tejedor Martínez 2009: 56-57;Forster, Borgwaldt, and Neef 2012: 122;Stange andNübling 2014: 1983;Bednarczyk 2015: 10;Jensen, Hougaard and Levisen 2019: 2). In all those genres, interjections feature abundantly, and their usage is crucial for the transmission of a message and ultimately for successful communication. Given this, drama plays, emails, WhatsApp messages, and comic strips can all be regarded as particularly suitable for the study of interjections and their behaviors. They offer rich and easily searchable corpora in which the grammatical profile of the category of interjections can be examined holistically, in both qualitative and quantitative terms. Such corpora are especially useful when examining the syntactic properties of interjections (Ameka 1992(Ameka , 2006Nübling 2004;Ameka and Wilkins 2006;Stange and Nübling 2014;Stange 2016).
The present paper aims to advance the study of interjections in isiXhosa by examining the syntactic behavior of the interjectional category from both a qualitative and quantitative perspective, as attested in one of the written-language genres where interjections are particularly abundant -comic strips. This research will be conducted within a prototype-driven approach to interjections (Ameka 1992(Ameka , 2006Nübling 2004;Ameka and Wilkins 2006;Stange and Nübling 2014;Stange 2016) and will answer the following research question: To what degree do interjections, found in the comic genre in isiXhosa, comply with syntactic features postulated for an interjectional prototype in scholarly literature -in particular, its syntactic extra-systematicity?
To achieve its objective, the article will be divided into four parts. In the next section (Section 2), we will introduce the framework underlying our research and the method with which the empirical corpus study was conducted. In Section 3, we will describe the behavioral profile of Xhosa interjections, emerging from our database. In Section 4, this profile will be evaluated within the adopted framework and the answer to the research question will be provided. Lastly, in Section 4, conclusions will be drawn.

Methodology: theoretical framework and method
As indicated by its name, the fundamental concept in a prototype-driven approach to interjections is the prototype. The prototype of an interjection has been proposed in light of the features regularly exhibited by interjections across languages, as well as those features' saliency, i.e. the most conspicuous differentiation from other word classes (Ameka 1992(Ameka , 2006Nübling 2004;Ameka and Wilkins 2006;Stange and Nübling 2014;Stange 2016).
Even though the prototype of an interjection is a central concept in our approach, it cannot be equaled with the interjectional category itself. As defended by cognitive linguistics, the prototype structures the category but neither encapsulates nor represents it exhaustively (Croft 2003;Evans and Green 2006;Janda 2015). Crucially, instantiations of a category attested across languages may exhibit varying degrees of compliance with the prototype, being canonical (where more features are met) or non-canonical (where fewer features are met). The category envisaged in its totality is itself represented as a radial network of members that emanate from the center (canonical members) to the periphery (non-canonical members). The center is exemplified by the prototype, while at the periphery, the category overlaps with other categorial networks. Although both canonical and non-canonical members belong to the same categorytheir membership status is uneven. As a result, categories become gradient phenomena, and their borders are viewed as fuzzy. 3 Given the gradient nature of an interjectional network, language-specific interjections need not comply with the syntactic extra-systematicity associated with the prototype; or they may only do so to a certain extent. Indeed, most syntactic properties viewed as inherent to the interjectional prototype are violated across languages, with greater or lesser regularity.
Interjections may appear sentence-internally or even clause-internally. The phonological separation of interjections from the remaining parts of an utterance (or sentence) may be minimal with no pause or extra-sentential contour being present (Ameka and Wilkins 2006: 6;Nordgren 2015: 45). In extreme instances, interjections can be clitisized to other lexemes, thus forming a single prosodic word (Ameka and Wilkins 2006: 6).
Given the syntactic features associated with the prototype of interjections, and their violations, the method adopted in the empirical research on which this paper draws is the following: for each interjectional token, we study the properties that could reveal the extent of compliance or defiance of the syntactic prototype.
We primarily analyze whether each interjectional token constitutes an independent utterance or whether it is dependent -forming part of a larger utterance. 4 In the former case, we examine whether it is isolated constituting a unique utterance in a given turn, or whether it appears as an initial, internal, or final utterance in a sequence of utterances in a turn. In the latter case -i.e. if an interjection constitutes a part of an utterance, we study: its prosodic relationship to the adjacent elements indicated by the various punctuation signs (comma, ellipsis, apostrophe) or their absence (juxtaposition); the lexical class of the adjacent element(s) (e.g. whether it is a noun (vocative or non-vocative), pronoun, verb, adverb, connective, or clitic particle); and, as previously, the position of the interjectional lexeme (i.e. initial, final, or internal). Additionally, we have examined whether an interjectional component -be it independent or dependentappears singly (one interjection is used); whether it is replicated (two, three, or more identical interjections are used); or whether it is used in combination with another interjection forming a chain (two or more different interjections are used). At the end of our empirical study, we compile a searchable multi-parameter database consisting of 1849 tokens.
In accordance with the prototype-driven approach, all semantic types of interjections irrespective of the degree of their interjectionality (Nübling 2004;Stange and Nübling 2014;Stange 2016: 17-18) have been included in our database and research, i.e.: (a) expressive (emotive and cognitive) interjections, which encode the emotional, sensorial, and mental states of the speaker (e.g. o, yho, or awu used to indicate surprise, anger, or fear); (b) conative interjections, which express wishes, convey commands, and function as attention getters (e.g. shh(h) and sssshhh to request silence); and (c) phatic interjections, which specify the attitude of the speaker towards the conversation and discourse (by initiating, sustaining, or terminating communication) and convey gratitude (enkosi), apology (uxolo), and salutations during welcoming (hey(i), molo, bhota(ni)) or leave-taking (sala(ni) kakuhle). 5 Similarly, we have included in our database both primary (e.g. oh and yhu) and secondary interjections (e.g. Nkos(i y)am). 6 Overall, all interjectional lexemes present in our database feature in the comprehensive set of interjections that has been compiled by Andrason and Dlali (forthcoming). 7 The corpus from which the tokens composing our database have been extracted comprises the complete selections of half-a-page or one-page comic strips that were published monthly -with some interruptions -in

Evidence
Interjections constitute a highly prevalent feature of the comics analyzed for the purpose of this study. As mentioned above, there are 1849 interjections in the entire corpus, which gives an average of 7,5 interjections per single comic -whose length, as explained, oscillates between half a page or one page.
Although not all frames contain interjections, most of them do. In extreme cases, a single frame comprises more than ten interjectional items. In various frames, interjections constitute the main bulk of the utterances employed (1) or are the only words used (2). Frames that contain interjections as their unique elements may be monologic (they reproduce speech of only one person (3)) or dialogic (two or more speakers intervene (2 and 4)). Overall, interjections are exploited more commonly for dialogic (5 and 6) than monologic purposes ( 6 Regarding the distinction between primary and secondary interjections consult Ameka (1992Ameka ( : 111, 2006, Norrick (2009), Stange and Nübling (2014: 1982-1983, Meinard (2015), and Stange (2016: 9, 19). 7 This set was compiled by reviewing all isiXhosa dictionaries and grammars and by conducting original field work. 8 The relevant interjections will be marked in bold throughout the article. Each example will be accompanied by a reference indicating the year and the months of publication. The capital letters A, B, C … used in some examples stand for different speakers/characters intervening in a story. It should be noted that we will not translate the interjectional items as the equivalence between most isiXhosa interjections and their English rendering is elusive. Moreover, since our study is concerned with syntax, the exact meaning of the interjections is less relevant.
Mostly, interjections (more than 66%) are used as independent, fully fledged, and selfcontained utterances. 9 Such interjectional utterances do not constitute elliptical or shortened versions of longer expressions. In a majority of cases, an independent interjection appears as the only utterance in a turn (7,8,9). 10 We will refer to that position as isolated. To be exact, 74% of the independent interjections are isolated, which equals nearly 49% of all the tokens. E-e-na D: Ukuwa ngendololwane? B: Eaayi. E: Imitya yezihlangu ezitsha? B: Hai maan. (1996/08) A smaller part of independent interjections, amounting to 26%, are not isolated but appear adjacent to other utterances in a turn (10-13). In nearly 70% of such cases, a non-isolated independent interjection occupies an initial position, featuring at the beginning of a turn (10 and 11). A final position is significantly less common being attested in only 16% (12). An internal position is the least frequent -14% of the non-isolated independent interjections are placed between two utterances (13). Independent interjections that are not isolated tend to be separated from the adjacent utterances by the exclamation mark (see (10), (11), and (13) above). Much less commonly, the separation is achieved by an interrogation mark (14). Exceptionally, and in general only when used in a final position, a non-isolated independent interjection is separated from the other utterances by a full stop (see (12)  Even though independent interjections predominate, the use of interjections as parts of utterances is not exceptional. This type of usage is found in 34% of all the cases. In such instances -which will be referred to as dependent -an interjection is not the only component of an utterance. Rather, it forms that utterance together with other words: verbs, nouns, pronouns, adverbs, connectives, particles, etc. As was the case with independent interjections, dependent interjections typically occupy an utterance-initial position. Specifically, 93% of dependent interjections are found at the beginning of an utterance (16 and 17). In contrast, the placement of dependent interjections in an utterance-final position is less common, amounting to 4% (18 and 19). Dependent interjections that occupy an internal position, being thus intercalated between other components of the utterance, are even rarer. They constitute 3% of all dependent uses. In some cases, a single utterance may host both initial and final interjections (20 A considerable number of utterance-initial dependent interjections is followed by a vocative noun (21 and 22). In isiXhosa, vocative nouns lack pre-prefixes, contrary to non-vocative forms used in affirmative as well as in certain types of negative constructions. Compare the vocative noun ntwana '(boy) child!' in (21) 12 with its non-vocative form umntwana; or the proper name Sticks used vocatively in (22) with its non-vocative variant uSticks. The presence of an utterance-initial interjection before 2 nd person pronouns -which allow for the same vocative reading (see nina 'you (pl.)' in (23)) -is also attested, being nevertheless less common than is the case of vocative nouns. Utterance-initial interjections may also be found before non-vocative (pro)nominal forms such as pre-prefixed nouns (i-R-20 '(the) 20 rand' in (24)) or pronouns other than the those referring to the 2 nd person singular or plural (mna 'I' in (25)). They may also be followed by verbs (ndiyaqonda 'I understand' in (26)), particles (ke in (27) and bo in (28)), adverbs (see 61 below), and other categories, e.g. copulatives (ngabafanas 'it is the young men (the Bafanas) again' in (29)), locative demonstrative copulatives (nantsi in (30)), ideophones (nqo 'straight' in (31)), and connectives (kodwa 'but, however' in (32)). As far as verbal inflectional categories are concerned, no restrictions are observed, for instance interjections are used with the perfect, the present, or the future tense. Similarly, utterance-final interjections accompany a variety of lexical classes such as: nouns, either vocative (mama 'mom' in (33)) or non-vocative (indiri 'brain' in (34)), 13 verbs (yonakele 'it is broken' in (35) and ndicela 'I ask' in (36)), or adverbs (see example (19) above), as well as more complex constructions, e.g. a prepositional phrase with a suffix pronoun (ngokwa(m) '(by) myself' in (37)). Dependent interjections are never integrated into the syntactic structure of the adjacent clause(s) with which they form an utterance. That is, they are not projected by the clause's predicate, nor do they function as adjuncts. They also fail to modify or complement the structural components of the clause, whether arguments or adjuncts (see examples 16-37 above).
The phonological relationship between interjections and the adjacent elements can only be inferred indirectly through punctuation used in writing. Dependent interjections (e.g., utteranceinitial (38) and utterance-final (39)) are separated from the remaining parts of the utterance by a comma in 37% of the possible cases. This overtly indicates the phonological disjunction of the interjection.
(38) Nkos' am, andazi (1992/01) (39) Niyahleka, hi? (1985/05) However, interjections may also be juxtaposed to the adjacent elements without any punctuation marks being used (40-43). This notation suggests that the interjection is phonologically conjunctive to the rest of a hosting utterance, or that a potential separation between the interjection and its utterance is minimal. 14 The use of juxtaposition is overall slightly more common (55%) than the presence of any overt orthographic "separatrix". It is attested with utterance-initial (40-42), as well as utterance-final interjections (43) Juxtaposition is particularly common in cases where the interjection precedes a vocative noun, e.g. ntwana '(boy) child!' (44) and Bombie (45), or the 2 nd person pronouns, e.g. wena 'you' (46) (see also examples 37, 38). Indeed, an interjection and a vocative (whether noun or pronoun) are almost never separated by a comma or ellipsis (see also (21,22,23)). However, a comma typically separates the sequence composed of an interjection and a vocative (or 2 nd person pronoun) from the remaining part of the utterance (44-45) (see also (22) above). Juxtaposition is even more regular when the interjection is followed by a particle (47-48). That is, the interjection and the particle are never separated by a comma or ellipsis (see also (27-28) above). Other lexical classes may also appear without punctuation signs that would overtly separate them from interjections, even though their orthographic separation is more typical. Those lexical classes are non-vocative nouns and pronouns (e.g. mna 'I' in (49)), verbs (e.g. ayaphukangaka 'it is not broken' in (50) and ukhabe 'you've kicked' in (51)), as well as ideophones (see example 31, above). Juxtaposition is also a common feature in cases where two different interjections are accumulated. Although the initial interjection in such chains is often separated from the subsequent one by a comma, the latter interjection tends to be linked to the remaining parts of the utterance through juxtaposition (52-54): Much less commonly (8%), a dependent interjection and the adjacent part of the utterancewhether the following (55-57) or the preceding one (58-59) -are linked via ellipsis. The use of ellipsis in writing suggests a great extent of phonological disjunction -significantly larger than was the case with a comma. The presence of ellipsis after utterance-initial interjections is more common than its use in front of utterance-final interjections. Ellipsis is also used to separate utterance-internal interjections from the left part of the utterance (60). In general, the vast majority of interjections (73%) are singletons. That is, they are used "monolexemically" -only one interjection is present (64-69). Interjectional singletons may function both as independent utterances (64-67) and as parts of utterances (68-69). They are also found in all positions. For instance, mono-lexemic independent interjections appear in an initial (64), final (65-66), and internal (see example (13) above) position, also allowing for isolated uses (67). Similarly, mono-lexemic dependent interjections are found utterance-initially (68), finally (69), and internally (see example (60) above). Although mono-lexemic interjections prevail, complex interjections are not uncommon. They are found in nearly 27% of all the cases. The main bulk of complex interjections involve the replication of one basic interjectional lexeme. Most often, an interjection is repeated twice, thus yielding cases of reduplication (70-74). The reduplicated interjections may belong to a larger utterance (70-71). They may also constitute the only elements of an utterance (72). Alternatively, they may, each, form an independent utterance, being separated -from one other as well as from the other utterances in a turn -by an exclamation sign (73-74). Much less frequently, an interjectional lexeme is replicated three times (75-77). Triplicated interjections may be located within a larger utterance or they may constitute the only elements of an utterance (75). Interjectional triplets may also function as three consecutive utterances, regularly separated by exclamation signs (76-77). Exceptionally -and in the analyzed corpus, only as part of larger multi-interjectional complexes -interjections may be repeated four or five times (see examples (86) and (88)

Discussion
The evidence presented in Section 3 demonstrates that interjections in isiXhosa often conform to the interjectional prototype as far as their syntax is concerned: (a) Interjections tend to function holophrastically, forming independent, complete, and selfcontained utterances.
Interjections frequently resist syntactic integration. This is evident for independent interjections and also often holds true for the interjections that form parts of utterances. Dependent interjections are never integrated into the structure of the adjacent clause with which they form an utterance. They do not constitute structural components of that clause, being, for instance, projected by its verb. They are neither governed by the verb's arguments and adjuncts, nor do they modify and complement the clause's structural elements. In various cases, dependent interjections do not enter into grammatical relationships with any words employed in the utterance, thus failing to yield broadly understood constructions. (c) Interjections exhibit a considerable degree of linear peripherality. In non-isolated uses, whether as independent utterances or as parts of larger utterances, interjections overwhelmingly occupy peripheral positions, typically the initial one. (d) Interjections are often separated topographically from the remaining parts of the sentence by a comma or (less commonly) ellipsis. This punctuation suggests a phonological disjunction of an interjection from the adjacent elements, and a likely interpretation of the interjectional segment as a separate prosodic unit. 17 However, the above-mentioned compliance with the interjectional prototype is not absolute, and several violations are attested: (a) Interjections often form parts of larger utterances.

(b)
Interjections are able to maintain syntactic relationships with grammatical elements at a phrasal and sentential level. On the one hand, interjections enter into constructions. Specifically, (i) interjections form phrases with vocative nouns and 2 nd person pronouns; (b) interjections fall under the scope of, or are restricted by, clitic particles; and (c) interjections yield interjectional chains, either by being replicated or by co-occurring with other interjections. (c) Interjections -both independent and dependent -may sporadically occupy internal positions. They are either intercalated between two utterances or placed utteranceinternally. (d) Interjections are often topographically joined to the remaining parts of an utterance. Overall, the use of juxtaposition is slightly more common than the use of all separatrices jointly. In rare instances, interjections can be linked to other lexemes by an apostrophe. The commonness of juxtaposition and the presence of an apostrophe-type linking suggest a high degree of phonological conjunction between interjections and the subsequent elements, allowing, in some cases, for the interpretation of these two segments as a single prosodic word.
The results of our empirical study demonstrate that the canonicity of Xhosa interjections is not uniform. On the one hand, interjections may exhibit a canonical profile. In such cases, they comply with the cross-linguistic interjectional prototype by being independent, non-integrated, linearly peripheral, and phonologically disjunctive. On the other hand, the canonicity of interjections may be more limited. Interjections can transgress prototypical features by being (more) dependent, integrated, linearly central, and phonologically conjunctive. Accordingly, the extra-systematicity of interjections fails to be uniform. Interjections oscillate between high levels of extra-systematicity (typical of canonical interjections) and low levels thereof (typical of non-canonical interjections). Consequently, rather than forming a homogeneous set, the category of interjections in isiXhosa is internally complex and diversified (cf. Nübling 2004: 11-16;Norrick 2009: 889), containing members of distinct canonicity and, thus, extrasystematicity degrees. The radical generalizations regarding the syntactic oddity of interjections occasionally found in isiXhosa scholarship (Louw and Jubase 1963: 255;Schonstein Pinnock 1994: 196;Oosthuisen 2016: 357) should therefore be regarded with more caution.
Additionally, the present research confirms certain phenomena and tendencies identified previously in scholarship, either by nuancing them further or by providing their robust empirical foundation. First, our study corroborates the view of the systemic relevance of interjections in isiXhosa grammar. While the high type-frequency of the interjectional category has been established by Andrason and Dlali (forthcoming), who have identified nearly 350 fully grammaticalized and/or lexicalized interjections in isiXhosa, our corpus demonstrates similarly impressive token-frequency. 18 At least in the comic genre, interjections are equally frequent as verbs and nouns -the most common lexemes in isiXhosa. In some comic strips, the occurrence of interjections even exceeds that of all the other lexical classes combined. Second, our research shows that violations of interjectional holophrasticity (i.e. instances where interjections are dependent) and non-constructionality (i.e. instances where interjections are used with vocatives and 2 nd person pronouns) are most commonly attested with conative and phatic interjections. This corroborates the less orthodox character of these two types of interjections if compared to 18 Regarding the distinction between type and token frequency, consult Hopper and Traugott (2003).
On the one hand, we agree that the interjectional prototype is essentially extra-systematic from a syntactic perspective, as proposed in scholarly literature. On the other hand, we observe a number of violations, which are, again, fully congruent with those identified previously in scholarship. In harmony with the prototype approach, interjections exhibiting such violations should not be denied categorial membership. Rather, they should be interpreted as occupying a peripheral position in the radial network of the interjectional category.

Conclusion
This paper offered an empirical, corpus-driven examination of the syntactic properties of interjections in isiXhosa -the first attempt of this in isiXhosa scholarship. The analysis demonstrates that isiXhosa interjections often comply with the prototype of interjections postulated in typological literature, thus exhibiting an extra-systematic profile. However, their canonicity -and thus extra-systematicity -is not universal. In various instances, interjections exhibit less canonical properties, being inversely more systematic from a syntactic perspective. As a result, the category of interjections in isiXhosa contains members of varying degrees of canonicity and extra-systematicity. Overall, interjections appear as one of the most frequent components of the discourses analyzed in this article, which suggests their crucial importance for communication generally, and in isiXhosa specifically.

Examples cited
(