11.07.2015 Views

Interjections in Dutch: a corpus-based approach

Interjections in Dutch: a corpus-based approach

Interjections in Dutch: a corpus-based approach

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong><strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>: a <strong>corpus</strong>-<strong>based</strong> <strong>approach</strong>Carla Schelfhout, Peter-Arno Coppen, Nelleke Oostdijk, Frans van der Slik;Radboud University NijmegenAbstractIn this paper, the distribution of <strong>in</strong>terjections over the clause is studied with<strong>in</strong> the context of a topologicalframework. For this study authentic data were extracted from a <strong>corpus</strong> compris<strong>in</strong>g one million words ofwritten material and 174,000 words of spoken material. To expla<strong>in</strong> this distribution we study the<strong>in</strong>fluence of the text type <strong>in</strong> which an <strong>in</strong>terjection is found, the length of the <strong>in</strong>terjection and its function.All three factors are shown to play a role. This confirms and amplifies hypotheses that hitherto had goneuntested.Samenvatt<strong>in</strong>gDit artikel bestudeert de distributie van <strong>in</strong>terjecties over de z<strong>in</strong> <strong>in</strong> termen van een topologisch model.Hierbij wordt gebruik gemaakt van authentieke data, die voorkwamen <strong>in</strong> een <strong>corpus</strong> van één miljoenwoorden geschreven materiaal en 174.000 woorden gesproken materiaal. Om deze distributie teverklaren, bestuderen we de <strong>in</strong>vloed van het teksttype waar<strong>in</strong> de <strong>in</strong>terjectie voorkomt, de lengte van de<strong>in</strong>terjectie en de functie van <strong>in</strong>terjecties. Alle drie factoren blijken een rol te spelen. Deze resultatenvormen een bevestig<strong>in</strong>g van en aanvull<strong>in</strong>g op totnogtoe onbewezen hypotheses.1 Introduction<strong>Interjections</strong> are undoubtedly the most-neglected word class <strong>in</strong> studies of <strong>Dutch</strong>,probably due to the prevail<strong>in</strong>g view that they are peripheral to the language. Theauthoritative descriptive grammar of <strong>Dutch</strong>, Algemene Nederlandse Spraakkunst(ANS, ‘General Grammar of <strong>Dutch</strong>’, Haeseryn et al. 1997), for example, states that<strong>in</strong>terjections are ‘words that are usually or always outside the grammatical structure ofthe sentence’ (translation ours, CS et al.).The ANS divides the rather broad class of <strong>in</strong>terjections <strong>in</strong>to three groups. First, itmakes a division between <strong>in</strong>terjections that have a mean<strong>in</strong>g and <strong>in</strong>terjections thatmerely imitate a sound, like miauw ‘meow’, wam ‘wham’, tsjoeketsjoek ‘choo choo’.<strong>Interjections</strong> with a mean<strong>in</strong>g are subdivided <strong>in</strong>to those that necessarily expressemotion, like swearwords, au ‘ouch’ or ocharme ‘oh, poor him’, and those that do not.This group is further divided <strong>in</strong>to announcements like foei ‘shame on you’ and ja ‘yes’,orders like ho ‘stop’ and toe (nou) ‘come on’, questions like hè ‘right’ and nietwaar‘isn’t it’, and social formulas like goedemorgen ‘good morn<strong>in</strong>g’, pardon ‘pardon me’and proost ‘cheers’.Although <strong>in</strong>terjections are assumed to be outside the grammatical structure of thesentence, they occur l<strong>in</strong>early with<strong>in</strong> this grammatical structure of the sentence. Thefew studies of <strong>in</strong>terjections, such as Van den Toorn (1968), Brummel (1978),Haegeman (1984), De Vriendt (1992) and Romijn (1998), assume that they can occur


Chapter 4-Calmost anywhere, though probably not with<strong>in</strong> major constituents. 1 However, thesestudies are either restricted to a few specific <strong>in</strong>terjections or <strong>based</strong> on <strong>in</strong>tuition, ratherthan on authentic data.Is their assumption correct? And if <strong>in</strong>terjections can <strong>in</strong>deed occur <strong>in</strong> any structuralposition, are they evenly distributed over these positions, or are certa<strong>in</strong> positionspreferred over others? Are there any differences <strong>in</strong> this respect between the text types<strong>in</strong> which <strong>in</strong>terjections are used? Do various types of <strong>in</strong>terjection (various <strong>in</strong> length,word or function) show different distributions? The answers to these questions arerelevant both for the descriptive study of <strong>Dutch</strong> and for a comparison of <strong>in</strong>terjections toother constructions which also are outside the grammatical structure of the sentencebut occur l<strong>in</strong>early with<strong>in</strong> the sentence, as for <strong>in</strong>stance vocatives or disjunctcomment<strong>in</strong>g clauses.We have carried out a <strong>corpus</strong>-<strong>based</strong> study <strong>in</strong>to the distribution of <strong>in</strong>terjections toanswer these questions. The distribution was studied at the level of the clause, ratherthan the utterance. This study was <strong>based</strong> on a <strong>corpus</strong> which conta<strong>in</strong>s both written andspoken material. For practical reasons, we considered only <strong>in</strong>terjections with<strong>in</strong> theboundaries of the utterance; <strong>in</strong>terjections <strong>in</strong> utterance-<strong>in</strong>itial or utterance-f<strong>in</strong>al positionwere excluded. We will show (1) that <strong>in</strong>terjections are not evenly distributed over theclause, and (2) how the length of the <strong>in</strong>terjections and the text type <strong>in</strong> which they occur<strong>in</strong>fluence their distribution. F<strong>in</strong>ally we compared <strong>in</strong>terjections with another type of<strong>in</strong>terrupt<strong>in</strong>g constructions, parentheticals, to determ<strong>in</strong>e whether function <strong>in</strong>fluencestheir distribution.2 The <strong>corpus</strong> materialThere are several considerations for the design of a <strong>corpus</strong> for a study of thedistribution of <strong>in</strong>terjections. The <strong>corpus</strong> must be large enough to provide sufficientmaterial for a statistically reliable analysis, yet small enough to keep its compilationand analysis feasible. It must comprise enough different text types to berepresentative of more than one language variety and, of course, it must be suitable tocheck our hypotheses.We expect to f<strong>in</strong>d differences <strong>in</strong> the use of <strong>in</strong>terjections <strong>in</strong> spoken and writtenlanguage. In particular, we expect <strong>in</strong>terjections <strong>in</strong> spoken language to be morefrequent, show a wider distribution and greater variation <strong>in</strong> types than <strong>in</strong> writtenlanguage. For written language, we expect a difference between text types whichsomehow reflect spoken language (<strong>in</strong>terviews, the dialogues <strong>in</strong> fiction texts) and texttypes which do not. In spoken language, it is conceivable that differences appear1 Except Van den Toorn (1968), who claims that <strong>in</strong>terjections almost always precede or follow theclause.


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>between public, often prepared text types like lectures, and private, spontaneous texttypes like private conversations. In spoken language, also differences betweenmonologues and dialogues can be expected. On the basis of these considerations, wecompiled a <strong>corpus</strong> with the follow<strong>in</strong>g design: 2Table 1: The <strong>corpus</strong> designWritten text type Number ofwordsSpoken text type Number ofwordsNovel 255,503 Lecture 62,810Short story 255,653 News 36,143Interview 126,376 Sports commentary 4,209News 123,140 Interview 7,101Essay 127,122 Private conversation 63,883Scientific writ<strong>in</strong>g 125,846Total 1,013,640 Total 174,146The written <strong>corpus</strong> was taken from the Internet. The l<strong>in</strong>guistic quality of materialfound on the Internet varies greatly, but the text types selected for the <strong>corpus</strong> designusually comprise carefully written material. The spoken <strong>corpus</strong> was taken from theSpoken <strong>Dutch</strong> Corpus (Corpus Gesproken Nederlands, CGN; Oostdijk 2000). Thewritten <strong>corpus</strong> was annotated by means of the same tagger that was used for theCGN, so that both parts had comparable annotation. All utterances conta<strong>in</strong><strong>in</strong>g at leastone <strong>in</strong>terjection were extracted from the tagged <strong>corpus</strong> with an automatic searchprogram. Instances <strong>in</strong> utterance-<strong>in</strong>itial or utterance-f<strong>in</strong>al position were discarded. If anutterance conta<strong>in</strong>ed more than one <strong>in</strong>terjection, each <strong>in</strong>stance was analyzedseparately, so that the total number of <strong>in</strong>stances is greater than the total number ofutterances. 3 As it is very difficult to decide objectively when two <strong>in</strong>terjections are usedadjacent to each other and when there is one multiword <strong>in</strong>terjection, we consider astr<strong>in</strong>g of <strong>in</strong>terjections, like ‘ja ja’ yes yes, to be one <strong>in</strong>terjection <strong>in</strong> all cases. In all, thereare 939 <strong>in</strong>stances, distributed over the text types as follows:2 In fact, this design was an <strong>in</strong>termediate stage <strong>in</strong> the compilation of a <strong>corpus</strong> with a spoken componentof 500,000 words, carefully divided over the spoken text types. However, at the moment that theresearch was carried out, the CGN had not yet been completed, so that the number of words availablefor each text type differs. As there is still a balance between prepared material on the one hand andspontaneous material on the other hand, we feel this should not bias our results.3 For practical reasons, uh was not regarded as <strong>in</strong>terjection <strong>in</strong> the spoken material.


Chapter 4-CTable 2: Distribution of <strong>in</strong>stances over the text types with a normalization per hundred thousand wordstext type # <strong>in</strong>terjections # <strong>in</strong>terjectionsper 100,000wordsNovel 56 22WrittenmaterialShort story 122 48Interview 31 25News 1 1Essay 13 10Scientific writ<strong>in</strong>g 1 1Written total 224 22SpokenmaterialLecture 130 207News 3 8Sports commentary 11 261Interview 70 986Private conversation 501 784Spoken total 715 411Total 939 79As expected, <strong>in</strong>terjections occur more frequently <strong>in</strong> spoken than <strong>in</strong> writtenlanguage. In written language, they occur more often <strong>in</strong> text types which reflectspoken language, viz. novels, short stories and <strong>in</strong>terviews. In spoken language, theyoccur less often <strong>in</strong> monologues (lectures, news, sports commentaries) than <strong>in</strong>dialogues (<strong>in</strong>terviews, private conversations). This all agrees with the expectations.The total number of different <strong>in</strong>terjections (types) is 86, 49 of which are used only once(hapax legomena). The most frequent <strong>in</strong>terjections are listed <strong>in</strong> Table 3.Table 3: <strong>Interjections</strong> occurr<strong>in</strong>g 10 times or more <strong>in</strong> the <strong>corpus</strong>Interjection Translation Frequencyja yes 490hè right 101nee no 69hoor really 51nou ja well 22nou well 18verdomme damn 15


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>ach oh well 12In order to study the distribution of <strong>in</strong>terjections over the clause, we need ananalysis model for the clause. This will be presented <strong>in</strong> the next section.3 Analysis modelTo <strong>in</strong>vestigate the positions at which <strong>in</strong>terjections occur, we follow the standardtopological model of <strong>Dutch</strong> traditional grammar as described <strong>in</strong> the ANS. This model isrelatively theory-neutral, so that the results can be transposed to several types ofsyntactic analysis. In this model, the clause is organized around its verbal positions,V1 and V2, with the middle field MI <strong>in</strong> between. Topicalized elements appear <strong>in</strong> TOP,extraposed elements are placed <strong>in</strong> EX. Dislocated elements occur either clause<strong>in</strong>itially(LD) or clause-f<strong>in</strong>ally (RD).• LD: the Left Dislocation field. This is the position for left-dislocated elements.• TOP: the topicalization field, which is the canonical position for subjects andtopicalized elements.• V1: the first verb field. In ma<strong>in</strong> clauses this field conta<strong>in</strong>s the f<strong>in</strong>ite verb, <strong>in</strong>subord<strong>in</strong>ate clauses, the subord<strong>in</strong>ator.• MI: the middle field, between the two verbal fields.• V2: the verbal cluster field; it conta<strong>in</strong>s all non-f<strong>in</strong>ite verbal elements <strong>in</strong> a ma<strong>in</strong>clause and all verbal elements <strong>in</strong> a subord<strong>in</strong>ate clause.• EX: the extraposition field, for extraposed elements.• RD: the Right Dislocation field, for right-dislocated elements.As def<strong>in</strong>ed, only the middle and extraposition fields can conta<strong>in</strong> more than onemajor constituent, i.e. clause-level constituents. These major constituents can becomposed of m<strong>in</strong>or constituents, for <strong>in</strong>stance an adverbial PP is composed of apreposition and a noun phrase, but it is the PP as a whole which fulfills a role at clauselevel. Table 4 illustrates the topological analysis.Table 4: Examples of analyses of clauses <strong>in</strong> a topological modelLD TOP V1 MI V2 EX RDJan,JohndiehimIkIkancanhebhaveik welI PRTdie manthat manschieten!shootgezienseenmet de hoedwith the hat


Chapter 4-Cdifferences between them are significant or not; we set α to .05. The result looks like atree structure <strong>in</strong> which the root is the entire set of data and the leaves are subgroups.The example <strong>in</strong> Figure 1 reflects the distribution over the clause of <strong>in</strong>terjections <strong>in</strong>written and spoken material.


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>Figure 1: The distribution of the clause of <strong>in</strong>terjections <strong>in</strong> spoken and written materialIn Figure 1 the root conta<strong>in</strong>s all <strong>in</strong>put data, which are divided over two leaves. Thepercentages given with the number of <strong>in</strong>stances at each position apply to one leaf, notto the root. Percentages with<strong>in</strong> one leaf always add up to 100. However, thepercentages between parentheses at the lowest row of any leaf <strong>in</strong>dicate thepercentage of the root. The difference between leaves is determ<strong>in</strong>ed from thedifferences between the percentages, to account for differences <strong>in</strong> group size. InFigure 1, the difference between spoken and written material is found to be significant(p = 0.01, chi-square = 21.2, df = 9). This seems to be the result of many smallerdifferences between the frequencies of occurrence <strong>in</strong> certa<strong>in</strong> positions.The leaves are then analyzed as new roots to determ<strong>in</strong>e whether there also aredifferences among the distributions of <strong>in</strong>terjections <strong>in</strong> the various text types with<strong>in</strong> thespoken or written material. For written material, there is no significant subdivision to bemade. For spoken material, there is a significant difference between privateconversations on the one hand and lectures, <strong>in</strong>terviews and sports commentaries onthe other hand (p < 0.005, chi-square = 32.1, df = 9). 6 The differences are large <strong>in</strong> thepositions between clauses and with<strong>in</strong> the middle field.The <strong>in</strong>stances <strong>in</strong> a transparent position are disregarded s<strong>in</strong>ce they cannot beassigned exactly. In addition, the three <strong>in</strong>stances <strong>in</strong> the category ‘spoken, news’ areexcluded because we wanted to exam<strong>in</strong>e differences <strong>in</strong> distribution depend<strong>in</strong>g on thetext type <strong>in</strong> which an <strong>in</strong>terjection occurs. Three <strong>in</strong>stances <strong>in</strong> a given category are<strong>in</strong>sufficient for conclusions. Therefore, the total number of <strong>in</strong>stances at the top of thetree is 909, not 939. For mean<strong>in</strong>gful results, we often had to disregard <strong>in</strong>stances thatoccurred only sporadically and whose classification <strong>in</strong> a certa<strong>in</strong> group thereforeseemed merely accidental. These cases are <strong>in</strong>dicated at the relevant places <strong>in</strong> thediscussion.5 ResultsThe first hypothesis to check is the standard assumption <strong>in</strong> the literature that<strong>in</strong>terjections can occur <strong>in</strong> all positions, but not with<strong>in</strong> major constituents. Relatedquestions are whether there are preferred positions for <strong>in</strong>terjections, whether theprosody of the clause <strong>in</strong>fluences the distribution of <strong>in</strong>terjections over the clause andwhether various types of <strong>in</strong>terjection behave differently. This section presents relevantdata and discusses their <strong>in</strong>itial implications; an extensive discussion of rema<strong>in</strong><strong>in</strong>gissues appears <strong>in</strong> Section 6.6 We used abbreviations for the text types <strong>in</strong> Figure 1; SP = private conversation, SL = lecture, SI =<strong>in</strong>terviews and SS = sports commentary.


Chapter 4-C1.1 The distribution of <strong>in</strong>terjections over the clauseTable 7 summarizes the distribution of <strong>in</strong>terjections over the positions <strong>in</strong> the clause <strong>in</strong>spoken and written material. 7 Apart from the hypothesis that <strong>in</strong>terjections can occur <strong>in</strong>almost any position except with<strong>in</strong> major constituents, no hypotheses can be derivedfrom the literature about which positions are preferred or avoided. We will study thedistribution of <strong>in</strong>terjections <strong>in</strong> the topological framework as discussed <strong>in</strong> Section 3: e.g.LD means ‘with<strong>in</strong> the left dislocation field’, LD-TOP means ‘between the leftdislocation field and the topicalization field’, TOP means ‘with<strong>in</strong> the topicalization field’and so on. The sign # <strong>in</strong>dicates the between-clause position and TRANSPARENTmeans that the position cannot be determ<strong>in</strong>ed exactly.Table 7: The distribution of <strong>in</strong>terjections over the clause <strong>in</strong> the topological framework <strong>in</strong> written andspoken materialPosition #<strong>in</strong>terjections<strong>in</strong> writtenmaterial%<strong>in</strong>terjections<strong>in</strong> writtenmaterial#<strong>in</strong>terjections<strong>in</strong> spokenmaterial%<strong>in</strong>terjections<strong>in</strong> spokenmaterialLD 1 0.4 6 0.8LD-TOP 15 6.7 23 3.2TOP 13 5.8 45 6.3TOP-V1 0 0.0 11 1.5V1-MI 15 6.7 16 2.2MI 22 9.8 92 12.9MI-V2 2 0.9 6 0.8V2 0 0.0 0 0.0V2-EX 1 0.4 6 0.8EX 3 1.3 13 1.8EX-RD 0 0.0 0 0.0RD 0 0.0 0 0.0# 143 63.8 479 67.0TRANSPARENT 9 4.0 18 2.5Totals 224 100 715 1007 In Section 3 we suggested that LD, TOP and RD could only conta<strong>in</strong> one major constituent. S<strong>in</strong>ce<strong>in</strong>terjections are supposed not to <strong>in</strong>terrupt major constituents, this would imply that EX and MI are theonly topological fields <strong>in</strong> which <strong>in</strong>terjections can occur. LD and TOP are also used, however. In thesecases <strong>in</strong>terjections are used when people repeat themselves, correct themselves or want to bridge thegap while they th<strong>in</strong>k about the cont<strong>in</strong>uation.


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>The position #, clause boundary, is most frequent: about 65% of the <strong>in</strong>terjectionsoccur <strong>in</strong> this position. The middle field, MI, is the most frequent clause-<strong>in</strong>ternalposition, roughly 10% of the cases. The positions LD-TOP, TOP and V1-MI accountfor about 5% of the cases each, and the rema<strong>in</strong><strong>in</strong>g positions are only used <strong>in</strong> a fewpercent of the cases. Some positions, viz. V2, EX-RD and RD are not used at all. Theabsence of <strong>in</strong>stances <strong>in</strong> EX-RD and RD is easily expla<strong>in</strong>ed, s<strong>in</strong>ce these positions arerarely available; the right-dislocation field RD is rare. The absence of <strong>in</strong>stances <strong>in</strong> thesecond verb field, V2, cannot be expla<strong>in</strong>ed <strong>in</strong> the same way; this field is relativelyfrequent and can conta<strong>in</strong> more than one word. The absence of <strong>in</strong>terjections hereapparently <strong>in</strong>dicates that the coherence of the elements <strong>in</strong> the verbal cluster is toostrong for <strong>in</strong>terruption by an <strong>in</strong>terjection.Although the distribution of written and spoken material only differs a few percentper position, the difference is significant (p = 0.01, chi-square 21.5, df = 9; cf. Figure1). The distribution of <strong>in</strong>terjections <strong>in</strong> the spoken material is significantly differentbetween private conversations on the one hand and sports commentary, lectures and<strong>in</strong>terviews on the other hand (p < 0.005, chi-square = 32.1, df = 9). 8 This contrastsuggests a stylistic difference between public and private speech.It appears that the general assumption about major constituents is correct:<strong>in</strong>terjections rarely <strong>in</strong>terrupt them. Of the 939 <strong>in</strong>stances we found, only 25 <strong>in</strong>terrupteda major constituent, all <strong>in</strong> the spoken material. Another 80 <strong>in</strong>stances, 17 <strong>in</strong> the writtenmaterial and 63 <strong>in</strong> the spoken material, occurred <strong>in</strong> situations of self-correction,repetition of a word or a str<strong>in</strong>g of words, restarts and the like. In these situations it isdebatable whether they are <strong>in</strong>terrupt<strong>in</strong>g a major constituent or not, but <strong>in</strong> our view theyare not.Although Table 7 suggests that <strong>in</strong>terjections show preferences for certa<strong>in</strong>positions, we have to bear <strong>in</strong> m<strong>in</strong>d that this would only be true if all positions wereequally available for <strong>in</strong>terruption by an <strong>in</strong>terjection. This is obviously not the case,however. A certa<strong>in</strong> topological position is only available for an <strong>in</strong>terjection when thisparticular topological field is occupied, e.g. <strong>in</strong> the clause ‘hij is verdwaald’ he got lost,as exemplified <strong>in</strong> Table 5, the positions EX, EX-RD and RD are not available. It seemsstraightforward that the field RD will be used less frequently than the field MI, andhence that <strong>in</strong>terjections can occur less often <strong>in</strong> the former field than <strong>in</strong> the latter. Toaccount for this difference, the frequencies of <strong>in</strong>terjections <strong>in</strong> certa<strong>in</strong> fields should bedivided by the frequencies of use of those fields.However, to our knowledge, there is no large <strong>corpus</strong> of written or spoken <strong>Dutch</strong>analyzed accord<strong>in</strong>g to the topological framework, so the relative frequencies of8 We removed the three <strong>in</strong>stances from the category news items s<strong>in</strong>ce this number is too low to baseconclusions about the entire category on it. Besides, we also removed the transparent <strong>in</strong>stances.


Chapter 4-Ctopological fields are unknown. Therefore we must be careful with conclusions abouthigh frequencies of <strong>in</strong>terjections at certa<strong>in</strong> positions. They might reflect a largeravailability of these positions rather than a prosodic, stylistic or other reason to preferthese positions. However, this caveat holds both for written and for spoken materialand for all types of <strong>in</strong>terjection; therefore conclusions about the differences betweengroups of <strong>in</strong>stances can be drawn safely from this position analysis.1.2 Does the prosody of the clause <strong>in</strong>fluence the distribution of <strong>in</strong>terjections?The <strong>in</strong>fluence of prosody on the distribution of <strong>in</strong>terjections is difficult to predict, s<strong>in</strong>cemany factors play a role <strong>in</strong> the prosody of a clause. Unfortunately the prosody cannotbe observed <strong>in</strong> the data; obviously the written material is not prosodically annotated,and only a small proportion of the CGN received a prosodic annotation. We restrictedour analysis to one factor only: it seems likely that the length of an <strong>in</strong>terjection could<strong>in</strong>fluence the positions at which it can occur. We expect that the longer an <strong>in</strong>terjectionis, the more difficult it becomes to <strong>in</strong>tegrate <strong>in</strong>to the <strong>in</strong>tonation contour of the clause.The number of syllables proved a good <strong>in</strong>dicator: we found a significant difference (p


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>Most <strong>in</strong>stances are monosyllabic <strong>in</strong>terjections; there are 757 monosyllabic<strong>in</strong>terjections (most of which are ja or hè, cf. Table 3), while there are 153 polysyllabic<strong>in</strong>terjections. The ma<strong>in</strong> difference between monosyllabic and polysyllabic <strong>in</strong>terjectionsis that monosyllabic <strong>in</strong>terjections occur more often at clause boundaries and less oftenat the positions with<strong>in</strong> and preced<strong>in</strong>g the middle field, MI and V1-MI. The differencesbecome clearer when we divide the groups of monosyllabic and polysyllabic<strong>in</strong>terjections by written or spoken text.Table 9: The distribution over the clause of monosyllabic <strong>in</strong>terjections <strong>in</strong> spoken and written materialPosition In spoken material In written material# % # %LD 5 0.81 1 0.71LD-TOP 21 3.41 13 9.22TOP 42 6.82 12 8.51TOP-V1 8 1.30 0 0.00V1-MI 15 2.44 0 0.00MI 81 13.15 8 5.67MI-V2 6 0.97 0 0.00V2-EX 5 0.81 0 0.00EX 12 1.95 2 1.42# 421 68.34 105 74.47Totals 616 100 141 100Table 10: The distribution over the clause of polysyllabic <strong>in</strong>terjections <strong>in</strong> spoken and written materialPosition In spoken material In written material# % # %LD 1 1.23 0 0.00LD-TOP 2 2.47 2 2.78TOP 3 3.70 1 1.33TOP-V1 3 3.70 0 0.00V1-MI 1 1.23 15 20.83MI 11 13.58 14 19.44MI-V2 0 0.00 2 2.78V2-EX 1 1.23 1 1.39EX 1 1.23 1 1.39# 58 71.60 36 50.00Totals 616 100 141 100


Chapter 4-CCompar<strong>in</strong>g the distribution of monosyllabic and polysyllabic <strong>in</strong>terjections <strong>in</strong> writtenmaterial, polysyllabic <strong>in</strong>terjections show a stronger preference for the positions V1–MIand MI, while monosyllabic <strong>in</strong>terjections are more frequent between clauses. Thedifferences between monosyllabic and polysyllabic <strong>in</strong>terjections <strong>in</strong> spoken material,however, are now much narrower. Apparently the differences <strong>in</strong> position preferencebetween monosyllabic and polysyllabic <strong>in</strong>terjections on the whole are due almostexclusively to the differences <strong>in</strong> written material. This is remarkable, s<strong>in</strong>ce one wouldexpect the <strong>in</strong>fluence of prosody to be greater <strong>in</strong> spoken material than <strong>in</strong> writtenmaterial. A closer exam<strong>in</strong>ation reveals that 42 of the 72 polysyllabic <strong>in</strong>terjections <strong>in</strong> thewritten material are swearwords. An alternative explanation for the significantdifference between monosyllabic and polysyllabic <strong>in</strong>terjections <strong>in</strong> written materialmight therefore be that swearwords have a distribution which is significantly differentfrom the distribution of other types of <strong>in</strong>terjection. We will return to this issue <strong>in</strong> Section6.1.3 The distribution of various types of <strong>in</strong>terjection over the various text typesWhen we look at the preferences of each type of <strong>in</strong>terjection for a certa<strong>in</strong> text type, wesee that there is a significant difference (p < 0.005, chi-square = 260, df = 17) betweenspoken and written material. That holds at least for <strong>in</strong>terjections that occur oftenenough <strong>in</strong> the material to yield statistically significant results; these are only the 18types of <strong>in</strong>terjection that occur five times or more <strong>in</strong> the data. 10Table 11: The distribution of 18 types of <strong>in</strong>terjection over spoken and written materialType of Translation In spoken In written<strong>in</strong>terjectionmaterial material# % # %ach oh well 3 0.45 9 5.17godverdomme goddamn 0 0.00 5 2.87goed okay 5 0.75 0 0.00hè right 100 15.06 1 0.57hoor really 38 5.72 13 7.47ja yes 425 64.01 64 36.78ja goed yes okay 9 1.36 0 0.00ja hoor oh yes 0 0.00 7 4.02ja nou def<strong>in</strong>itely 5 0.75 0 0.0010 These 18 types together cover 841 <strong>in</strong>stances. In order to study the distribution over text types, weremoved the 3 <strong>in</strong>stances <strong>in</strong> spoken news; this number was too small to draw reliable conclusions fromthese <strong>in</strong>stances.


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>nee no 31 4.67 38 21.84nee hoor oh no 1 0.15 5 2.87nou well 16 2.41 2 1.15nou ja well 17 2.56 5 2.87oh oh 4 0.60 2 1.15pardon pardon me 3 0.45 1 0.57sorry sorry 7 1.05 1 0.57verdomme damn 0 0.00 15 8.62verdorie darn 0 0.00 6 3.45Totals 664 100 174 100The ma<strong>in</strong> differences are that hè and ja have a strong preference for spokenmaterial, whereas nee occurs much more frequently <strong>in</strong> written material. verdorie,godverdomme and verdomme occur exclusively <strong>in</strong> written material. The difference <strong>in</strong>use over the text types between ja and nee is remarkable, s<strong>in</strong>ce <strong>in</strong>tuitively thesewords are each other’s counterparts, so that one would expect an equal distribution. Apossible explanation, suggested by a look at the <strong>in</strong>stances, is that ja is often used witha non-affirmative function <strong>in</strong> spoken material, such as back channel or as a filler orplaceholder, a function which nee is less likely to fulfil. An example from the spoken<strong>corpus</strong> is given <strong>in</strong> 2.2. en zo ben ik uh ja eigenlijk al op deze school beland waar ik (...)and so am I uh yes actually already on this school ended_up where I‘and that's how I, well, ended up at this school where I...’S<strong>in</strong>ce it is not necessary to prevent an <strong>in</strong>terruption by the conversation partner <strong>in</strong>written material, this function is rare <strong>in</strong> writ<strong>in</strong>g; the only exception is fiction dialogues or<strong>in</strong>terviews. This could expla<strong>in</strong> why the differences between the frequencies of ja andnee are smaller <strong>in</strong> written material than <strong>in</strong> spoken material. This explanation seemseven more likely when we exam<strong>in</strong>e the spoken material, <strong>in</strong> Table 12. 11Table 12: The distribution of 14 types of <strong>in</strong>terjection over four spoken text typesInterjection PrivateconversationLecture,<strong>in</strong>terview, sportscommentaries# % # %ach 2 0.43 1 0.5111 The types which do not occur <strong>in</strong> the spoken material are not depicted <strong>in</strong> Table 12.


Chapter 4-Cgoed 3 0.64 2 1.02hè 32 6.85 68 34.52hoor 29 6.21 9 4.57ja 337 72.16 88 44.67ja goed 9 1.93 0 0.00ja nou 2 0.43 3 1.52nee 22 4.71 9 4.57nee hoor 1 0.21 0 0.00nou 14 3.00 2 1.02nou ja 10 2.14 7 3.55oh 4 0.86 0 0.00pardon 0 0.00 3 1.52sorry 2 0.43 5 2.54Totals 467 100 197 100The distribution of <strong>in</strong>terjections <strong>in</strong> the spoken material reveals two groups: theprivate conversations on the one hand and lectures, <strong>in</strong>terviews and sportscommentaries on the other (p < 0.005, chi-square = 112, df = 13). The ma<strong>in</strong> differenceis that <strong>in</strong> private conversations the word ja occurs more frequently than <strong>in</strong> publicmaterial. The use of ja as a filler or placeholder is more often necessary <strong>in</strong> privateconversations than <strong>in</strong> sports commentaries or lectures, where an <strong>in</strong>terruption by theconversation partner is unlikely.6 DiscussionThe results presented <strong>in</strong> Section 5 confirm the standard assumption <strong>in</strong> the literature:<strong>in</strong>terjections can occur <strong>in</strong> almost all structural positions, but they rarely <strong>in</strong>terrupt amajor constituent. Their distribution differs between spoken and written material, and<strong>in</strong> the spoken material it also differs between public monologues on the one hand andprivate dialogues on the other hand. This probably reflects a stylistic differencebetween those text types. Various <strong>in</strong>terjections show clear preferences for text types,presumably also due to stylistic differences. Whether the length of an <strong>in</strong>terjection<strong>in</strong>fluences its distribution rema<strong>in</strong>s unclear; at the end of this section we will presentadditional research <strong>in</strong>to this question. First, we will discuss two other results of Section5: the very frequent use of the position between clauses, #, and the almost entireabsence of swearwords <strong>in</strong> spoken language.


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>1.1 The preference of <strong>in</strong>terjections for the position between clausesThe frequency of the between-clause position # is so high, 65%, that it can hardly beexpla<strong>in</strong>ed by the frequent availability of this position. Apparently there is a reason toprefer <strong>in</strong>terjections at positions between clauses over positions with<strong>in</strong> clauses. Theexplanation may be related to the prosody of the clause and the extra-grammaticalnature of <strong>in</strong>terjections. Clause boundaries seem more suitable for extra-grammaticalconstructions than clause-<strong>in</strong>ternal positions; besides, these boundaries often show aprosodic pause, which may make it easier to <strong>in</strong>terject a clause-external element. Thefact that the second preference for <strong>in</strong>terjections is the position with<strong>in</strong> the middle fieldseems to falsify this hypothesis, but many of these <strong>in</strong>stances <strong>in</strong>dicate hesitation, selfcorrectionand the like. An example from the spoken <strong>corpus</strong> is given <strong>in</strong> 3:3. dat zal over uh zo ongeveer uh nou_ja zesenhalve m<strong>in</strong>uut zijn.that will over uh so about uh well six_and_a_half m<strong>in</strong>ute be‘That will be <strong>in</strong>, well, about six m<strong>in</strong>utes and a half.’The need to keep the turn while th<strong>in</strong>k<strong>in</strong>g or to correct the previous word appliesmost frequently with content words that carry new <strong>in</strong>formation, which often occur <strong>in</strong> themiddle field. This goes for spoken language as well as for fiction dialogues. Inaddition, <strong>in</strong>terjections can occur as <strong>in</strong>tensifiers <strong>in</strong> the written material, as <strong>in</strong> 4, which isderived from the <strong>corpus</strong>:4. (...)de geëxecuteerde (...) karakteriserend als die stoere Belg ja zelfsthe executed characteriz<strong>in</strong>g as that tough Belgian yes eventoesprekend met (...)address<strong>in</strong>g with‘... characteriz<strong>in</strong>g the executed one as that tough Belgian, yes even address<strong>in</strong>g himwith...’Thus there is a functional explanation for the occurrences <strong>in</strong> the middle field,which need not exclude the explanation for the preference for the between-clauseposition. Of course, further research <strong>in</strong>to this hypothesis is necessary.1.2 SwearwordsAt first sight it seems surpris<strong>in</strong>g that swearwords occur more often <strong>in</strong> written than <strong>in</strong>spoken material. The most likely explanation is that all speakers <strong>in</strong> the CGN knew thattheir speech was be<strong>in</strong>g recorded. This might have made them more careful <strong>in</strong> thechoice of their words, a well-known drawback of the legal obligation to ask people’s


Chapter 4-Cpermission to record their speech <strong>in</strong> advance. This observation is re<strong>in</strong>forced by thefact that common taboo words like fuck ‘fuck’, kut ‘cunt’, shit ‘shit’ do not occur <strong>in</strong> thedata anywhere. The three swearwords verdorie ‘darn’, godverdomme ‘goddamn’ andverdomme ‘damn’ have become a bit old-fashioned and less shock<strong>in</strong>g than they oncewere. It makes them suitable for use <strong>in</strong> fiction dialogue, which is the most importantsource of these <strong>in</strong>stances.1.3 Prosodic and functional <strong>in</strong>fluence on the distribution of <strong>in</strong>terjectionsThe results presented <strong>in</strong> Section 5 suggest that the position of an <strong>in</strong>terjection isdeterm<strong>in</strong>ed by its length, among other factors. However, about 3/5 of the polysyllabic<strong>in</strong>terjections <strong>in</strong> the written material are swearwords. This makes it very difficult todecide whether it is really the length of an <strong>in</strong>terjection which plays a role here, whichwould imply <strong>in</strong>fluence of the prosody, or the type of <strong>in</strong>terjection, which would suggestthat the function of the <strong>in</strong>terjection plays a role. This issue cannot be resolved bydivid<strong>in</strong>g the group of polysyllabic <strong>in</strong>terjections <strong>in</strong>to two groups, viz. swearwords andother <strong>in</strong>terjections, s<strong>in</strong>ce too few <strong>in</strong>stances would rema<strong>in</strong> <strong>in</strong> each group to ga<strong>in</strong> reliableresults.An alternative <strong>approach</strong> is to classify the <strong>in</strong>stances on the basis of their mean<strong>in</strong>g,as argued <strong>in</strong> the ANS and summarized <strong>in</strong> Section 1. If each <strong>in</strong>stance <strong>in</strong> the <strong>corpus</strong> isclassified <strong>in</strong>to a group accord<strong>in</strong>g to its mean<strong>in</strong>g, we can test whether these groupshave different distributions. The first problem is that many <strong>in</strong>stances do not occur <strong>in</strong>their primary mean<strong>in</strong>g but are used with a discourse function. ja ‘yes’ often acts as aplaceholder, bridg<strong>in</strong>g the gap while the speaker is th<strong>in</strong>k<strong>in</strong>g, and pardon ‘sorry’ or sorry‘sorry’ do not ask for clarification or repetition as the ANS suggests, but <strong>in</strong>troduce aself-correction. The second problem is that some <strong>in</strong>terjections can be used <strong>in</strong> differentbut related ways; so the classification becomes rather subjective. The same problemapplies to classify<strong>in</strong>g <strong>in</strong>terjections not <strong>in</strong>cluded <strong>in</strong> ANS’ examples. The description isnot formal enough to classify new <strong>in</strong>stances objectively. Because of the serious risk ofsubjectivity, we did not apply this classification to our data. Consequently, we couldnot test whether a difference <strong>in</strong> mean<strong>in</strong>g of an <strong>in</strong>terjection is <strong>in</strong> any way related to adifferent distribution.To determ<strong>in</strong>e whether the difference between monosyllabic and polysyllabic<strong>in</strong>terjections <strong>in</strong> written material was due to their length or to the type of the polysyllabic<strong>in</strong>terjections, we compared <strong>in</strong>terjections with a different type of <strong>in</strong>terruptionconstruction. We chose the so-called parentheticals; constructions like lijkt me ‘itseems to me’, dacht ik ‘I thought’, meen ik ‘I th<strong>in</strong>k’. The function of these constructionsis to add a meta-comment to the surround<strong>in</strong>g clause, provid<strong>in</strong>g the <strong>in</strong>formation that itis just a personal op<strong>in</strong>ion. This function is clearly different from the function of


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong><strong>in</strong>terjections. What parentheticals have <strong>in</strong> common with <strong>in</strong>terjections, however, is thatthey are not a grammatical part of the clause; both of them <strong>in</strong>terrupt the clause (cf.Schelfhout et al. 2004). In addition, parentheticals consist of at least two words;consequently they are all at least two syllables long. Therefore we feel that acomparison of parentheticals and polysyllabic <strong>in</strong>terjections could shed more light onthe role of the function of an <strong>in</strong>terruption versus the role of the length of the<strong>in</strong>terruption <strong>in</strong> decid<strong>in</strong>g its position.We extracted the utterances conta<strong>in</strong><strong>in</strong>g one or more parentheticals from a <strong>corpus</strong>which is a superset of the <strong>corpus</strong> from which the <strong>in</strong>terjections were extracted. Thewritten part of the <strong>corpus</strong> is exactly the same, but the spoken part was extended toalmost half a million words. We found 271 parentheticals <strong>in</strong> this material. 12 These<strong>in</strong>stances were encoded <strong>in</strong> the same way as <strong>in</strong>terjections with respect to the text types<strong>in</strong> which they orig<strong>in</strong>ated, their position and their length. The swearwords wereseparated from the <strong>in</strong>terjections to be able to see with what type of <strong>in</strong>terruption theywould cluster. The comb<strong>in</strong>ed data were analyzed by AnswerTree.We checked whether the different types of <strong>in</strong>terruption resulted <strong>in</strong> a differentpattern of position preference. 13 As parentheticals always conta<strong>in</strong> at least twosyllables, <strong>in</strong>terjections consist<strong>in</strong>g of only one syllable were removed from the data tokeep the comparison fair. Parentheticals and swearwords together show a distributionsignificantly different from that of all other types of <strong>in</strong>terjections (p < 0.005, chi-square= 80.5, df = 8).Table 13: The distribution of <strong>in</strong>terjections, swearwords and parentheticals over the clausePosition <strong>Interjections</strong>without swearwordsSwearwords andparentheticals# % # %LD-TOP 3 2.75 7 2.31TOP 4 3.67 8 2.64TOP-V1 2 1.83 15 4.95V1-MI 2 1.83 59 19.47MI 13 11.93 108 35.64MI-V2 0 0.00 6 1.98V2-EX 2 1.83 9 2.97EX 2 1.83 7 2.31# 81 74.31 84 27.7212 For more <strong>in</strong>formation on the parentheticals, see Schelfhout, Coppen & Oostdijk (2003).13 Interruptions at a transparent position (36 <strong>in</strong>stances) were removed, as were <strong>in</strong>stances occurr<strong>in</strong>g atthe position RD, which was used <strong>in</strong> only 4 of 1210 <strong>in</strong>stances. The position LD was used only once <strong>in</strong>this restricted set; this <strong>in</strong>stance was removed for reasons of parsimony.


Chapter 4-CTotals 109 100 303 100The difference between the distribution of parentheticals and swearwords is at theborder of <strong>in</strong>significance (p = 0.05, chi-square = 15.3, df = 8).These differences suggest that the function of an <strong>in</strong>terruption can cause asignificant difference <strong>in</strong> the preference for a certa<strong>in</strong> position. More specifically,parentheticals and swearwords show a stronger preference for positions preced<strong>in</strong>gand with<strong>in</strong> the middle field, V1-MI and MI, and occur less frequently at a clauseboundary (#) <strong>in</strong> comparison to all other types of <strong>in</strong>terjection. But does a differentfunction expla<strong>in</strong> all the differences? When we classify the entire set of data by thelength of the <strong>in</strong>terruptions, 14 we obta<strong>in</strong> a significant result as well.Table 15: Distribution over the clause of <strong>in</strong>terruptions with a length of one, two or more syllablesPosition One syllable Two syllables More syllables# % # % # %LD 6 0.79 1 0.37 0 0.00LD-TOP 34 4.49 7 2.59 3 2.10TOP 54 7.13 10 3.70 2 1.40TOP-V1 8 1.06 7 2.59 10 6.99V1-MI 15 1.98 33 12.22 28 19.58MI 89 11.76 74 27.41 47 32.87MI-V2 6 0.79 3 1.11 3 2.10V2-EX 5 0.66 9 3.33 2 1.40EX 14 1.85 8 2.96 1 0.70# 526 69.48 118 43.70 47 32.87Totals 757 100 270 100 143 100Now there is a clear subdivision between <strong>in</strong>terruptions of one syllable, of twosyllables and of three and more syllables long (p < 0.005, chi-square = 212, df = 8).The categories which conta<strong>in</strong> polysyllabic <strong>in</strong>terruptions conta<strong>in</strong> a mixture of<strong>in</strong>terjections (exclud<strong>in</strong>g swearwords), swearwords and parentheticals; it looks like thetwo-syllable and more-syllable categories are not identifiable with ma<strong>in</strong>lyparentheticals or ma<strong>in</strong>ly <strong>in</strong>terjections, but really conta<strong>in</strong> both k<strong>in</strong>ds of structures. 15Therefore, it appears that both the length of an <strong>in</strong>terruption and its function <strong>in</strong>fluencethe distribution.14 We used five categories for the length of an <strong>in</strong>terruption. It can be 1 through 4 syllables long, or it isfive or more syllables long.15 Of course, the monosyllabic category does not conta<strong>in</strong> parentheticals.


Free <strong>in</strong>tercalations - <strong>Interjections</strong> <strong>in</strong> <strong>Dutch</strong>7 Conclusion<strong>Interjections</strong> can occur <strong>in</strong> all structural positions <strong>in</strong> the clause, but they cannot <strong>in</strong>terruptmajor constituents. Although we cannot be certa<strong>in</strong> without more <strong>in</strong>formation on therelative use of topological fields, <strong>in</strong>terjections seem to show preferences for certa<strong>in</strong>positions. The position between clauses seems to be the most favoured one, followedby the position with<strong>in</strong> the middle field. The explanation for these preferences isprobably a comb<strong>in</strong>ation of grammatical, functional and prosodic factors. The use of<strong>in</strong>terjections is clearly different <strong>in</strong> spoken and written language. In written material,<strong>in</strong>terjections are more frequent <strong>in</strong> text types which reflect spoken language use. Inspoken material, we also see a difference between the use of <strong>in</strong>terjections <strong>in</strong> privateand public text types. The difference can be seen both <strong>in</strong> the types of <strong>in</strong>terjection and<strong>in</strong> the positions they occupy. Those positions also seem to be <strong>in</strong>fluenced by the lengthof the <strong>in</strong>terjection: monosyllabic <strong>in</strong>terjections differ significantly from polysyllabic<strong>in</strong>terjections, especially <strong>in</strong> written material. A closer look reveals that polysyllabic<strong>in</strong>terjections <strong>in</strong> written material are ma<strong>in</strong>ly swearwords, which raises the questionwhether the length or the function of the <strong>in</strong>terjection causes the difference <strong>in</strong>distribution. A comparison between two functionally different types of <strong>in</strong>terruption,<strong>in</strong>terjections and parentheticals, suggests that both prosodic and functional factors<strong>in</strong>fluences their distribution.This study confirms and amplifies standard assumptions <strong>in</strong> the literature notpreviously tested with authentic material. It also raises issues for future research,amongst which the exact role of the function versus the prosody of <strong>in</strong>terjections <strong>in</strong>decid<strong>in</strong>g the distribution over the clause, their function <strong>in</strong> the discourse and the<strong>in</strong>teraction of these factors. Also the need for a <strong>corpus</strong> annotated accord<strong>in</strong>g to thetopological descriptive model is underl<strong>in</strong>ed once more, s<strong>in</strong>ce this is the only way toga<strong>in</strong> <strong>in</strong>sight <strong>in</strong>to the preferential distribution of <strong>in</strong>terjections. In view of the vary<strong>in</strong>gfrequency of <strong>in</strong>terjections over different text types and <strong>in</strong> different positions, more datawould be helpful for f<strong>in</strong>e-tun<strong>in</strong>g the analyses presented here, especially to determ<strong>in</strong>ewhether there is a difference <strong>in</strong> use between swearwords and other types of<strong>in</strong>terjection. More data could also be beneficial to determ<strong>in</strong>e whether there aredifferences between the various written text types, for which we have too few<strong>in</strong>stances at the moment. F<strong>in</strong>ally, the overall distribution shows that more than half the<strong>in</strong>stances occur <strong>in</strong> a s<strong>in</strong>gle position, between clauses. S<strong>in</strong>ce there are clauses ofmany different types, which can be coord<strong>in</strong>ated or subord<strong>in</strong>ated to each other, thevariation with<strong>in</strong> this group is large. It would be worthwhile to split up this group, for<strong>in</strong>stance <strong>in</strong>to positions between coord<strong>in</strong>ated and subord<strong>in</strong>ated clauses, or <strong>in</strong>topositions preced<strong>in</strong>g clauses which do or do not carry thematic roles, and see if thisprovides more <strong>in</strong>formation.


Chapter 4-CAcknowledgementsWe would like to thank Antal van den Bosch and Hans van Halteren for their help <strong>in</strong>tagg<strong>in</strong>g the written <strong>corpus</strong> and Bill Fletcher for proofread<strong>in</strong>g this paper. We are alsothankful to two anonymous reviewers and the editors of Neerlandistiek.nl for theircomments and helpful suggestions.ReferencesBrummel, G. (1978). Enkele opmerk<strong>in</strong>gen over <strong>in</strong>terjecties. In Berkel, A. (ed.),Proeven van Neerlandistiek aangeboden aan Prof. Dr. Albert A. Sassen, 155-175. Gron<strong>in</strong>gen: Nederlands <strong>in</strong>stituut.Haegeman, L. (1984). <strong>Interjections</strong> and Phrase Structure. In L<strong>in</strong>guistics 22: 1, 41-49.Haeseryn, W., K. Romijn, G. Geerts, J. de Rooij & M.C. van den Toorn, eds. (1997).Algemene Nederlandse Spraakkunst. Gron<strong>in</strong>gen / Deurne: Mart<strong>in</strong>us Nijhoff /Wolters Plantyn.Oostdijk, N. (2000). Build<strong>in</strong>g a <strong>corpus</strong> of Spoken <strong>Dutch</strong>. In Monachesi, P. (ed.),Computational L<strong>in</strong>guistics <strong>in</strong> the Netherlands: Selected Papers from the TenthCLIN Meet<strong>in</strong>g, 147-159. Utrecht: Universiteit Utrecht.Romijn, K. (1998). Eh: substitutie- en aarzel<strong>in</strong>gs<strong>in</strong>terjectie. In TABU 28: 2, 72-87.Schelfhout, C., P.-A. Coppen & N. Oostdijk (2003). Positions of parentheticals and<strong>in</strong>terjections: A <strong>corpus</strong>-<strong>based</strong> <strong>approach</strong>. In Cornips, L. & P. Fikkert (eds.),L<strong>in</strong>guistics <strong>in</strong> the Netherlands 2003, 155-166. Amsterdam: John Benjam<strong>in</strong>sPublish<strong>in</strong>g Company.——— (2004). F<strong>in</strong>ite comment clauses <strong>in</strong> <strong>Dutch</strong>: a <strong>corpus</strong>-<strong>based</strong> <strong>approach</strong>. In Journalof Germanic L<strong>in</strong>guistics 16: 4, 331-349.Van den Toorn, M.C. (1968). De <strong>in</strong>terjectie als woordsoort. In Hoogteijl<strong>in</strong>g, J. (ed.),Taalkunde <strong>in</strong> artikelen, 115-119. Gron<strong>in</strong>gen: Wolters-Noordhoff. Also pr<strong>in</strong>ted <strong>in</strong>De Nieuwe Taalgids 53 (1960), 260-264.De Vriendt, S. (1992). 'kom', 'kijk', 'zeg' als <strong>in</strong>terjectie. In Studia Neerlandica etGermanica, 513-520.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!