challenges in the linguistic description of romanian weak pronouns

25
Challenges in the linguistic description of Romanian weak pronouns UiT Norges Arktiske Universitet Ciprian-Virgil Gerstenberger [email protected] January 27, 2015 1 Introduction Despite undertakings such as (Yngve and Wasik, 2004) that try to put linguistics on a par with established hard sciences (e.g., biology, physics, and chemistry), linguistics is regarded as a soft science due to the difference in the degree of accuracy and objectivity, employment of controlled experiments, the production of testable prediction models and the use of mathematically exact ones. In the last decades, linguistics experienced a rapid change in research methods – from predominantly theoretical to more and more empirical – fact that is caused by the extremely fast development of digital data compilation and sharing. Yet, should a linguist working with large amounts of language data ignore the descriptions made by theoretical linguists? Or vice versa, should a linguist working theoretically ignore insights won by empiric work? Any theoretical linguist had to work with some language data, any empicially work- ing linguist has to have at least a simple theoretical framework in order to interpret and evaluate the empiric against it. Yet, factors such as the amount or the provenance of lan- guage data play a crucial part. Using Romanian weak pronouns as the phenomena under scrutiny, this essay tries to shed light on some general problems of language description. Besides strong personal pronouns, pronouns that can receive contrastive stress, Roma- nian possesses a set of weak pronouns, pronouns that can not be used with contrastive stress. These kind of pronouns are not specific to Romanian, as many languages feature 1

Upload: ciprianno

Post on 13-Sep-2015

5 views

Category:

Documents


2 download

DESCRIPTION

Compared to hard sciences (e.g., biology, physics, and chemistry), linguistics is regarded as a soft science due to the difference in the degree of accuracy and objectivity, employment of controlled experiments, the production of testable prediction and the use of mathematical models. In the last decades, linguistics experienced a crucial change in research methods – from pure theoretical to more and more empirical ones – fact that is caused by the extremely fast development of the internet, communication and digitaldata compilation. Yet, should a linguists working with large amounts of data in digital formal ignore the descriptions made by theoretical linguists? If not, what is the best way to combine the strength of theory with that of empiricism? Using Romanian cliticpronouns as the phenomena under scrutiny, this essay tries to shed light on some general problems of language description.

TRANSCRIPT

  • Challenges in the linguistic description ofRomanian weak pronouns

    UiT Norges Arktiske UniversitetCiprian-Virgil [email protected]

    January 27, 2015

    1 IntroductionDespite undertakings such as (Yngve and Wasik, 2004) that try to put linguistics on apar with established hard sciences (e.g., biology, physics, and chemistry), linguistics isregarded as a soft science due to the difference in the degree of accuracy and objectivity,employment of controlled experiments, the production of testable prediction models andthe use of mathematically exact ones.

    In the last decades, linguistics experienced a rapid change in research methods from predominantly theoretical to more and more empirical fact that is caused by theextremely fast development of digital data compilation and sharing. Yet, should a linguistworking with large amounts of language data ignore the descriptions made by theoreticallinguists? Or vice versa, should a linguist working theoretically ignore insights won byempiric work?

    Any theoretical linguist had to work with some language data, any empicially work-ing linguist has to have at least a simple theoretical framework in order to interpret andevaluate the empiric against it. Yet, factors such as the amount or the provenance of lan-guage data play a crucial part. Using Romanian weak pronouns as the phenomena underscrutiny, this essay tries to shed light on some general problems of language description.Besides strong personal pronouns, pronouns that can receive contrastive stress, Roma-nian possesses a set of weak pronouns, pronouns that can not be used with contrastivestress. These kind of pronouns are not specific to Romanian, as many languages feature

    1

  • them. In tradition literature, they are referred to as atonal pronouns, weak pronouns,or clitic pronouns. Yet, many other non-pronominal items are labeled clitics, too,(e.g., subjunctions, auxiliaries, adverbs), and for the sake of trying to be neutral, I willrefer to them as Romanian weak pronouns. For the same reason, I will try to use themore general term sandhi, the Sanskrit for joining, when other researchers would use thenotion of cliticization, cliticisation, or clitisation.

    First, I will show basic problems of the traditional concept the clitic-host pair. Then,I will point to the fact that trying to define the notion of clitic is almost as hard asto define whether an embryo is already a human being or not. What are the relevantfeatures for cliticness? is the crucial question.

    Further, I will review different approaches to the Romanian weak pronouns and willpoint to the sore spots of their argumentation. As will be demonstrated, different typesof ambiguities are the main source for their tenuous description and modeling. Subse-quently, I will provide the solution for the ambiguity issues: disambiguate wherever thisis needed. Then, I will reflect on my own approach and on the results as well as on themanner of communication between the different researchers that dealt with Romanianweak pronouns. I will conclude this essay with a possible interpretation of Karl Poppersfalsifiability demand.

    2 Clitic-host pairGenerally, clitics are items that can belong to any grammatical category with syntacticcharacteristics of a word, yet are phonologically bound to another word. This means theyshare both properties with words and with affixes. Conceptually, they can either leanfrom the left as proclitic or from the right as enclitic to a host. In the Frenchsentence Je taime. (eng. I love you.), the accusative pronoun t leans from the leftto its host, the verb aime: a clear case of proclitic. In the Latin expression SenatusPopulusque Romanus (eng. The Senate and Roman people) the conjunction que leansfrom the right to its host Populus: a clear case of enclitic.

    This neat description of clitics get blurred by a series of facts. As shown in Fig. 1,clitics occur often as a sequence, which can be a mix of different categories (subjunctions,negation, auxiliaries, adverbs). The first question is whether all clitics in such a cliticsequence lean the same way to the host or not. Lets have a closer look. In Romanian,both the auxiliary am in Ex. 1 and the weak pronoun le in Ex. 2 are regarded as clitics tothe verbal host. What is then the state in such a simple clitic sequence as in Ex. 3? Both

    2

  • Figure 1: A. Chereche, A Prosodic Analysis of Romanian Pronominal Clitics, 2014, p. 2, Ex.(1)

    weak pronouns, mi and le, are syllabic and non-stressed (even non-stressable). Does thewhole clitic sequence form a clitic-host pair with the main verb or each weak pronounindividually? The picture gets even more complicated if we take into account Ex. 4,where the non-syllabic form of the weak pronoun leans directly on the auxiliary, whichprovides the syllabicity of the syllable /leam/. Which is here the host, the auxiliary orthe verb?

    (1) Amhave_1.sg

    vzut.seen

    [Am vzut.]

    I have seen.(2) Le

    them_3.pl.accvd.see_1.sg.pres

    [Le vd.]

    I see them.(3) Mi

    me_1.sg.datlethem_3.pl.acc

    dai.give_2.sg.pres

    [Mi le dai.]

    You give them to me.(4) Le-

    them_3.pl.accamhave_1sg

    vzut.seen

    [Le-am vzut.]

    I have seen them.

    Since the sequence of mono- or non-syllabic items that behave in some respects alikebut not in all is a mix of different categories, a further question arises: which items reallydo belong to it and which not? The (non-stressed) subjunction s can be monosyllabic(see Ex. 5-7) and additionally, it can be the syllabic host for the non-syllabic weakpronoun form -mi, as in Ex. 6, on a par with the auxiliary am in Ex. 4. Yet, in specific

    3

  • contexts such as in Ex. 8, it surfaces as non-syllabic form. Does the subjunction s belongto the clitic sequence or not?

    (5) Vreauwant-1sg

    sthat

    mito me

    daigive-2sg

    cartea.book-def

    [Vreau s mi dai cartea.]

    I want you to give me the book.(6) Vreau

    want-1sgsthat

    -mito me

    daigive-2sg

    cartea.book-def

    [Vreau s-mi dai cartea.]

    I want you to give me the book.(7) Vreau

    want-1sgsthat

    aducibring-2sg

    cartea.book-def

    [Vreau s aduci cartea.]

    I want you to bring the book.(8) Vreau

    want-1sgs-that

    aducibring-2sg

    cartea.book-def

    [Vreau s-aduci cartea.]

    I want you to bring the book.

    There is no general agreement on the status of the verbal complex items as clitics.Some linguists claim that these items should be classified as affixes, some other claim thatthese should be classified as affixes only post-verbally. Again, other researchers claim thatthe items can be clitics pre-verbally, yet, thery are always clitics when following the mainverb, to mention only a couple a opinions. How is it possible that there is such a widerange of disagreement among linguists when it comes to label such language itemsas belonging to a category or another?

    Generally, we categorize things based on their properties, hence, we label something asbeing a specific entity based on its properties, properties that, in turn, we have to be ableto recognize. It is then clear that we can not speak about an entity, something belonging tothe realm of ontology, without refering to its recognizable properties, someting belongingto the realm of epistemology.

    That researchers disagree upon some category assignment in a specific domain is notsomething extaordinary neither something specific to linguistics, or, lets say, soft sciences.A famous example is assigning the category of planet to the celestial body Pluto. In 2006,the International Astronomical Union (IAU) established the three criteria to be met by acelestial body in our solar system to be classified as a planet. There, there is one criterionPluto doesnt meet in order to be labeled planet.1

    1http://en.wikipedia.org/wiki/IAU_definition_of_planet

    4

  • The crucial difference between linguistics and, say, astronomy is in the degree ofagreement upon the interpretation of properties, upon deciding which properties arerelevant for assigning labels to specific items. What is then relevance? Relevance is madeup by all neccessary and sufficient conditions in order to define a specific category. Isit at al possible to give a intesional definition of clitics at all? What about the notion ofword, one of the most central entities in linguistics? We have to keep in mind that thereis no clear-cut intesional definition of word that is unanimously accepted in the researchcommunity.

    3 To be or not to be a clitic or an affix or anAs described in the previous section, the concept of clitic-host pair is not quite compatiblewith that of clitic sequence, and one question related with the concept of sequence iswhether all items belonging to it have the same status. Besides something labeled cliticdoubling (see Ex. 17-27), Romanian features syllabic vs. non-syllabic weak pronounsurface forms that depend on the specific context of their occurence. This is perhapsthe reason why the term mesoclitic is not that often encountered in the literature.However, in the linguistic description of European Portuguese, this term seems to bemore fashionable. In the European Portuguese clause conquistar-se- (eng. it willbe conquered) the pronoun se is labeled as clitic while the present form of the verbto have, namely , that together with the infinitive form of the main verb form thesynthetic future tense, is regarded as an affix. Yet, a similar picture can be found inRomanian, especially in the nowaday not that often used inverse forms of declarative andinterrogative sentences: cuceri-se-va (eng. it will be conquered).

    To repeat, the concept of mesoclitic as a clitic leaning in the middle of a lexem isnot compatible with the original concept of clitic. Since it is a phenomenon describedon phonological level cliticization have to occur with a strong host, a phonological itemthat is stressed. Hence, if both the Romanian and the Portuguese se are clitics then bothhave to be enclitics. The only problem is to justify why an affix does not attach to itsverb immediately. The traditional concept of a lexem with stem and affixes is not validany longer because in these cases stem and affixes are interspersed with clitics.

    This is not the end of the story. In languages such as Udi, Pashto, or Degema, someitems labeled endoclitics are described as splitting the root into two pieces (e.g., Harris,2002), hence their domain of attachment is a part of the root. Not only that the conceptsof mesoclitic and endoclitic blur even more the clear clitic-host image, they even

    5

  • challenge the Lexical Integrity Hypothesis, too. This hypothesis represents the basicassumption that complex words can not be changed by syntactic rules.

    Why is so hard to agree upon a set of properties that clitics should have? Accordingto the clitic model put forward by (Klavans, 1995), some clitics can be regarded astransition stages of a grammaticalization process: lexical item ! clitic ! affix. Itis therefore understandable that clitics share properties both with lexical items and withaffixes. Although by far not with the same ethical and moral burden, the clitic dilemma issomehow comparable with the dilemma of deciding whether an embryo is already a humanbeing or not. Here too, the borderline between different stages is not that perspicuous.What are the relevant features that discriminate clitics from affixes?

    4 Relevant featuresAs with clitics in general, Romanian weak pronouns and other items labeled as cliticshave been and still are a highly debated the topic of research. As (Spencer andLus, 2013) points out, the checklist of rules to distinguish clitics from affixes in Englishprovided by (Zwicky and Pullum, 1983) has not proved too useful for applying it todifferent languages. Lets turn now to the relevant features that determined differentresearchers to label Romanian weak pronouns as clitics or as affixes.

    Compared with other clitic pronominal systems, the Romanian weak pronoun sys-tem convey both Latin and Slavic features, which is due to the Balkan Sprachbund(cf. Calude, 2001). Whether within Lexical Functional Grammar, Head-Driven PhraseStructure Grammar, Distributed Morphology, Optimality Theory, or other formalisms,the discussion whether they are clitics or affixes still has not come to an end. Amongmorphophonological models, (Barbu, 1999) and (Monachesi, 2005) claim that the Roma-nian weak pronouns are affixes, while (Popescu, 2000) and (Gerlach, 2002) label them asclitics.

    Working within Optimality Theory, (Chereche, 2014) delivers a prosodic analysis ofRomanian weak pronouns asserting that pronominals and auxiliaries are free prosodicwords in all environments, other than when they combine with each other: two pronomi-nals combine as internal clitics and auxiliaries combine with pronominals as affixal clitics(Chereche, 2014[p. 60]). The syntactic account in (Svescu Ciucivara, 2009) argues thatRomanian weak pronoun clusters behave differently from the point of view of internalorder in pre-verbal vs. post-verbal position, allegedly refuting the universality of thePerson Case Constraint. This account assumes that Romanian weak pronouns are the

    6

  • same type clitics both pre- and post-verbally. Like (Svescu Ciucivara, 2009), two othersyntactic account on the same topic, maintain that there is a difference between pre- andpost-verbal Romanian weak pronouns, yet the difference is not in the linearization behav-ior but in the type of clitic. The model in (Dobrovie-Sorin, 1999) introduces a distinctionbetween syntactic and phonological clitics in order to establish a common property toall clitic elements of different types (noun, adverb, verb). All Romanian weak pronounsare syntactic clitics but not all are phonological clitics: in pre-verbal position the weakpronouns may or may not be phonological clitics, while in post-verbal position they allshow obligatory phonological cliticization. In a similar vein, the description offered by(Klein, 2007) makes a clear distinction between pre-verbal and post-verbal Romanianweak pronouns as it is stated that we must distinguish between preverbal weak pro-nouns which encliticise, preverbal weak pronouns which procliticise, and preverbal weakpronouns which neither encliticise nor procliticise (postverbal weak pronouns are alwaysenclitic) (Klein, 2007[p. 13]).

    Models labeling Romanian weak pronouns as affixes do not consider crucial differ-ences between whatever is the normal behavior of an affix and the specificities of weakpronouns. Lets look closer at some arguments pro affix status. In (Monachesi, 2000), itis claimed that morphophonological idiosyncrasies are specific for affixes and thatsuch phenomena can be encountered also with the weak pronouns. The optional voweldeletion for weak pronouns ending with an such as m in front of a verb beginning withan unstressed a or o is compared with weak pronouns ending with an e such as ne, le,and te. The idiosyncrasy is that in the form se the final e deletes optionally in the samecontext as .

    The presentation of this argument is symptomatic for many other articles that focuson promoting a specific formal model thereby ignoring relevant data sets. First, thealleged idiosyncrasy is an optional form to a non-deleted yet non-syllabic variant (see Ex.9-16). Moreover, the form s- is obligatory for the non-optional sandhi such in s-a auzit(eng. has been hearable). I personally wouldnt say s-aude but only se-aude. Theform s-aude might be encountered in older texts or as a regionalism, perhaps in Moldova.Ergo: there is absolutely no idiosyncrasy there, obviously, the lack of knowledge aboutthe whole range of phenomena led (Monachesi, 2000) to draw a wrong conclusion.

    Another fact that (Monachesi, 2000) does not mention in this context is that alle-ending weak pronoun forms do enter optional sandhi in the mentioned contexts in aregular way. All weak pronoun forms in Ex. 9-12 are syllabic while those in Ex. 13-16 arenon-syllabic, which again puts the form s-aude (eng. is hearable/hears himself) on

    7

  • (9) ElHe

    mme

    aude.hears

    [El m aude.]

    He hears me.

    (10) ElHe

    teyou

    aude.hears

    [El te aude.]

    He hears you.

    (11) ElHe

    neus

    aude.hears

    [El ne aude.]

    He hears us.

    (12) ElHe

    sehimself

    aude.hears

    [El se aude.]

    He hears himself.

    (13) ElHe

    m-me

    aude.hears

    [El m-aude.]

    He hears me.

    (14) ElHe

    te-you

    aude.hears

    [El te-aude.]

    He hears you.

    (15) ElHe

    ne-us

    aude.hears

    [El ne-aude.]

    He hears us.

    (16) ElHe

    se-himself

    aude.hears

    [El se-aude.]

    He hears himself.

    a par with se-aude (eng. is hearable/hears himself) and te-aude (eng. hears you).It is worth mentioning that optional sandhi with different variants is not necessarilysomething uncommon: pe acolo, pe-acolo, and p-acolo (eng. around there) can all befound in spoken Romanian, and again, pe is syllabic while pe- and p- are non-syllabic, asthey are linked with a hyphen with the following word. Similar phenomena can be foundin many language, for instance, in German, which features variants such as andererseits,andrerseits, and anderseits for the eng. on the other hand.

    Even (Monachesi, 2000)s most convincing evidence in favor of the affixal statusof Romanian weak pronouns, the clitic doubling, loses its convincibility when we takeinto account further relevant language data. The clitic doubling, the fact that they [i.e.,Romanian weak pronouns] can co-occur with full complements behaving in those casesvirtually as agreement markers (Monachesi, 2000[p. 8]) is shown in Ex. 17, which is thesame as (Monachesi, 2000[p. 8, Ex. 24a.]). In a further example, (Monachesi, 2000) showsthat it is possible to have only the weak pronoun, hence without the full complement.

    Yet, (Monachesi, 2000) does not tell the whole, rather complicated, story of evenbanning the weak pronoun in specific contexts. The occurrence of weak pronoun and fullcomplement is controlled by a set of constraints over animacy vs. non-animacy, saliencevs. non-salience, given vs. new, topicalized (i.e., occurring on the first position in theclause) vs. non-topicalized, as is exemplified in Ex. 18-25. Due to the aforementionedconstraints, a co-occurrence of weak pronouns (aka agreement affixes) and full comple-ment is non-grammatical in Ex. 19, 21, and 23, while in Ex. 25 it is required. In contrast,

    8

  • the co-occurrence is optional if the complement is salient, as shown in Ex. 26 and 27. Tomy best knowledge, such kind of behavior is not typic for agreement affixes. A furthertotally atypic behavior for affixes is to occur both before and after the verb as a functionof verb mood: one and the same affix does not occur, say, as prefix in indicative and assuffix in imperative. The Latin affix ad fixum means to/in a fixed position.

    (17) MariaMaria

    iCL.DAT

    dgives

    preedinteluipresident-the

    una

    buchet.bouquet

    Maria gives the president a bouquet.

    (18) CiprianCiprian

    citetereads

    oacarte.book

    [Ciprian citete o carte.]

    Ciprian reads a book.(19) *Ciprian

    CiprianoPRON-CL-SG-FEM

    citetereads

    oacarte.book

    [*Ciprian o citete o carte.]

    Ciprian reads a book.(20) O

    acartebook

    citetereads

    Ciprian.Ciprian

    [O carte citete Ciprian.]

    Ciprian reads a book (not a newspaper).(21) *O

    acartebook

    oPRON-CL-SG-FEM

    citetereads

    Ciprian.Ciprian

    [*O carte o citete Ciprian.]

    Ciprian reads a book (not a newspaper).(22) Ciprian

    Cipriancitetereads

    cartea.the book

    [Ciprian citete cartea.]

    Ciprian reads the book.(23) *Ciprian

    CiprianoPRON-CL-SG-FEM

    citetereads

    cartea.the book

    [*Ciprian o citete cartea.]

    Ciprian reads the book.(24) Cartea

    the bookoPRON-CL-SG-FEM

    citetereads

    Ciprian.Ciprian

    [Cartea o citete Ciprian.]

    Ciprian reads the book.(25) *Cartea

    the bookcitetereads

    Ciprian.Ciprian

    [*Cartea citete Ciprian.]

    Ciprian reads the book.(26) El

    hemme-CL

    vedesees

    pePE

    mine.me

    [El m vede pe mine.]

    He sees me.

    9

  • (27) Elhe

    mme-CL

    vede.sees

    [El m vede.]

    He sees me.

    A similar case of leaving out or ignoring relevant data that not fit a specific modelor formalism is (Chereche, 2014)s description. (Chereche, 2014) makes the correct, butpartly too general and partly too specific, observation that [h]igh vowels are not allowedat the end of a Prosodic Word (Chereche, 2014[p. 55]) and declare Romanian weakpronouns and their combinations as prosodic words. Yet, this constraint applies to weakform of the copula verb and their combinations with weak pronouns, too (see Ex. 28).The form -s is the weak form of the copula snt2 (eng. are), whose vowel is a high vowel.

    (28) Crilebooks-DEF

    mito me

    -sare-CL

    dragi.dear

    [Crile mi-s dragi.]

    I love books.

    One interpersonally testable feature of Romanian weak pronouns is syllabicityvs. non-syllabicity, yet, in my humble opinion, even this issue is not handled properly.The main reason for this state is ambiguity on different levels, as explained below.

    As reproduced in Fig. 2, (Chereche, 2014) labels -prothetic forms such as i in (2)as long while those in (3) as short: note that the form i- in (3)a. is non-syllabic orthographically marked with a hyphen while the form i in (3)b. is syllabic thereis no hyphen linking the pronominal form to the adjacent item. A similar classificationuses also (Popescu, 2000[p. 775]) when partitioning the Romanian weak pronouns intofull and reduced forms: the -prothetic forms as well as the form in (3)b. are full formswhile the form in (3)a. is an instance of a reduced form in (Popescu, 2000). The samepartition as (Popescu, 2000) is offered also by (Guu-Romalo, 2008[p. 203]) by labeling(Popescu, 2000)s full forms as clitice libere (eng. free clitics) and Popescu, 2000)sreduced forms as clitice conjuncte (eng. bound clitics). The English version of(Guu-Romalo, 2008), (Dindelegan, 2013), keeps the same partition of Romanian weakpronouns into free and bound clitics (Dindelegan, 2013[p. 388]).

    As for the syntactic models offered by (Dobrovie-Sorin, 1999) and (Klein, 2007), theydont pay too much attention to the real phonologic value of a graphic string: what-ever is written with a hyphen and is not syllabic before the verb is a clitic for (Klein,2007) and a phonologic clitic for (Dobrovie-Sorin, 1999), and whatever is post-verbal isobligatorily clitic for (Klein, 2007) and obligatorily phonologic clitic for (Dobrovie-Sorin,

    2sunt in the orthography since 1993

    10

  • Figure 2: A. Chereche, A Prosodic Analysis of Romanian Pronominal Clitics, 2014, p. 52

    1999). What is the reason for these differences in the description of such a relativelytrivial phenomenon?

    5 Ambiguities in descriptionThe answer to the question about the discrepancies in the description of Romanian weakpronouns lies in various types of ambiguities. What is interesting is that, in strivingto achieve more clarity, some linguistic descriptions add even more ambiguous elementsto the phenomena described. How is this possible?

    Lets try to make out the different types of ambiguities that can occur or even haveoccurred in description of Romanian weak pronouns. The communication of researchbetween (not only) linguists is done mostly in written form (e.g., articles in journals,books). It is no mystery that there is no perfect mapping between speech and writing,therefore ambiguities between the two representations occur in some languages more,in other less frequently. The more ambiguities there are in a language, hence, also inthe language data presented in articles, the more additional information is needed fordisambiguation, especially in communication with non-native researchers.

    Homograph issuesAs a matter of fact, Romanian conveys a good deal of ambiguities both among Roma-nian weak pronouns and with other parts of speech. For instance, the string o can be3p.sg.fem.acc weak pronoun as in o vd (eng. I see her/it), indefinite article as in ocarte (eng. a book), or future particle in colloquial language as in o s citesc (eng. I

    11

  • Figure 3: M. Iliescu, Pentru o sistematizare a predrii pronumelui personal neaccentuat romnesc,Limba Romn 24, 1975

    will read). Further, i can be 3p.sg.masc.dat, 3p.sg.fem.dat, 3p.pl.masc.acc weak pro-noun forms, or the weak form of the copula verb a fi (eng. to be). Again, i can be3p.refl.dat weak pronoun form as in el i le cumpr (eng. he buys them for himself)or the conjunction i (eng. and) as in ea i el (eng. she and he). Moreover, s can be1p.sg.pres or 3p.pl.pres weak form of the copula to be as in crile mi-s dragi (eng. Ilove books), or the non-syllabic form of the reflexive weak pronouns in accusative as inel s-a splat (eng. he has washed himself), or the non-syllabic form of the subjunctions (eng. that) as in trebuie s-anun tirea (eng. I have to announce the news).

    Phoneme-grapheme issuesSince Romanian weak pronouns surface in different forms one of the biggest challenge isthe ambiguity between the phoneme and the grapheme string. For instance, the 1p.sg.datweak pronoun form is mi but the written string does not have always the same pronun-ciation (see Fig. 3). Although the orthography might give some clues for disambiguationit is not always reliable, as is explicated in the next paragraphs. The solution, a quitecumbersome one, is to disambiguate any and every written representation by adding thephonetic counterpart, as offered by (Iliescu, 1975) in Fig. 3.

    Hyphen issuesIn Romanian, all sandhi phenomena that involves desyllabification are marked by a hy-phen, yet, not all instances of hyphen are sandhi marking. As in other languages, the

    12

  • hyphen serves several purposes such as (1) marking compounds prietenia romno-bulgar(eng. the Romanian-Bulgarian friendship), (2) marking sandhi l-am vzut (eng. Ihave seen him), (3) marking any item in the verbal complex in post-verbal positionCumpr-i-le! (eng. Buy them (to yourself)!) as opposed to the pre-verbal positionas in i le cumperi. (eng. You buy them (to yourself).). Both i and le are syllabicboth in pre-verbal and in post-verbal position, hence the role of the both hyphens inCumpr-i-le! is solely to mark post-verbality.

    This tiny ambiguity in the use of hyphen obviously led (Dobrovie-Sorin, 1999) and(Klein, 2007) to the wrong conclusion that whatever weak pronoun comes after the verbhas to be a phonological clitic in (Dobrovie-Sorin, 1999)s description and clitic at allin (Klein, 2007)s description. An accurate examination shows that this is by no meanstrue because both models mix up syllabic and non-syllabic surface forms. In both Ex. 29and 30 te is syllabic despite the fact that the former is written without hyphen while thelater is written with hyphen. In Ex. 31, however, te is non-syllabic, therefore the sandhibetween te and acas is marked with a further hyphen, as the standard orthographyrequires. To repeat, (Dobrovie-Sorin, 1999) means by phonological clitic a reduced, i.e.,non-syllabic form, and (Klein, 2007) means the same by clitic. Now, how comes thatan already reduced form can further reduce by entering a sandhi with the subsequentitem? Or to put it in (Dobrovie-Sorin, 1999)s terms: if mi in Ex. 6 is phonological cliticto the subjunction s and at the same time syntactic clitic to the main verb dai for thepre-verbal configuration and if te in Ex. 31 is syntactic clitic and obligatory phonologicalclitic to the main verb du for the post-verbal configuration then how is it possible for teto be phonological clitic also to the adverb acas? A three-fold clitic-host relationship?

    (29) Teyourself

    ducicarry

    acas.home

    [Te duci acas.]

    You go home.(30) Du

    carry-teyourself

    acas!home

    [Du-te acas!]

    Go home!(31) Du

    carry-te-yourself

    acas!home

    [Du-te-acas!]

    Go home!

    I hope that these elaborations have been clear enough to illustrate the problems withthe function overloading of the hyphen in Romanian orthography.

    13

  • -Prothesis issuesAt the beginning of the previous problem description called Hyphen issues, I statedthat in Romanian all sandhi phenomena that involves desyllabification are marked by ahyphen. This is true only if we ignore the real purpose of in the -prothetic forms suchas mi, i, i, i. Referring to those forms, I keep using -prothetic because exactly thisis the purpose of the vowel /W/: to be syllabic prothesis to weak pronoun and weakcopula forms ending with a high vowel in specific configurations, namely as the only itemin the sequence.

    Despite the fact that the phenomenon of -prothesis has been observed and describedas early as 1976 by (Lombard, 1976), there are very few references to it and even less isthis observation taken into account in the analysis of Romanian weak pronouns. This isthe reason of the questionable classifications into long vs. short by Chereche, 2014, fullvs. reduced by (Popescu, 2000), or free vs. bound weak pronouns by (Guu-Romalo,2008) and (Dindelegan, 2013) described above.

    Again, the standard orthography is ambiguous once more: the inconsistency lies inusing the hyphen as desyllabification marker in all contexts involving Romanian weakpronouns but in contexts where the syllabic nucleus is provided by the -prothesis. Irre-spective of their individual syllabic support, both mi forms Ex. 32 and 33 have exactlythe same phonological shape, namely /mj/: mi /Wmj/ and s-mi /s@mj/. This shapesameness between the two orthographically different forms has been already pointed toby (Iliescu, 1975), again, a further article that is hardly taken into account in formaldescriptions of Romanian weak pronouns.

    (32) Vreauwant-1sg

    sthat

    mito me

    daigive-2sg

    cartea.book-def

    [Vreau s mi dai cartea.]

    I want you to give me the book.(33) Vreau

    want-1sgsthat

    -mito me

    daigive-2sg

    cartea.book-def

    [Vreau s-mi dai cartea.]

    I want you to give me the book.

    That the -prothesis can not be analyzed as a proper part of the weak pronouns isevident also from the fact that exactly the same vowel /W/ plays the same protheticrole in the weak forms of the copula a fi (eng. to be), as exemplified in Ex. 34 and 35.And again, this observation has been already made by (Lombard, 1976).

    (34) Eithey

    mito me

    -sare-CL

    dragi.dear

    [Ei mi-s dragi.]

    14

  • I like them.(35) Ei

    theysare-CL

    acas.at home

    [Ei s acas.]

    They are at home.

    Language description issuesAt the beginning of this section, I dared to make the bold claim that, in striving toachieve more clarity, some linguistic descriptions add even more ambiguous elements tothe phenomena described. Let me now show some facts that substantiate this claim.

    The task of this research is not to find and assess the quality of language data examplesthroughout the research articles on Romanian or even only on Romanian weak pronouns,yet I am bound to point that a thorough evaluation of examples found in the literature as has been done for Norwegian by (Engh, 2006) is badly needed. In extensive useof clitic forms, some copy&paste error or typo can slip in, such is the case with -miin (Soare, 2005[p. 49, Table 1]), assuming that the authors have a good command ofRomanian. With non-native speakers it is harder to assess whether, e.g., Pleca voi. in(Legendre, 2000[p. 13, Ex. 13Rb.]) is a typo or lack of knowledge. In (Legendre, 1999[p.9, Ex. 16Rb.]), the same author has exactly the same example with the correct spelling,namely Pleca-voi..

    Using the Leipzig Glossing Rules (LGR) cliticization by the equals sign = andsegmentable morphemes separated by a hyphen (Klein, 2007[p. 17]) , (Klein, 2007) seemsto avoid to use the hyphen as orthographic element of the object language, although notexplicitly stated. Yet, this leads to ambiguities such as Mercedes-ul (eng. the Mercedes)in (Klein, 2007[p. 96, Ex. 44b.]), which can be as well an orthographic element, since inspecific contexts the definite article is spelled with a hyphen. Moreover, although in mostcontexts the equals sign as cliticization marker seems to be used in a similar way as thehyphen in the standard orthography, in some contexts the standard orthography wouldhave been a more precise description. Fig. 4 depicts (Klein, 2007[p. 63, Ex. 14]) with theaddition of the correct orthographic string as the top line for comparison reasons. Whilethe first equals sign in the example i=l=a misleads the reader in suggesting that thereis some sandhi between i and l, the use of standard orthography is unambiguous in thisrespect for pre-verbal contexts as it spells i l-a.

    It is almost always a guessing exercise when it comes to the use of Leipzig Gloss-ing Rules in Romanian examples: is the hyphen part of the meta-language, hence ofthe glossing rules, or of the object language, i.e., of the Romanian orthography? In

    15

  • Figure 4: (Klein, 2007[p. 63, Ex.14]) enriched with the correct orthographic string as the top line

    Figure 5: A. Zwicky, On clitics, 1977, p. 20

    (Chereche, 2014[p. 52, Ex. 3a.]) reproduced in Fig. 2 , the hyphen is probably partof the orthography.

    One has to be wary of questionable language data even in frequently quoted articlessuch as (Zwicky, 1977). The only two Romanian examples reproduced here in Fig. 5 are both spelled the wrong way, even in the orthography from 1928 or 1976. The correctspelling is biatul i va da-o and biatul i-o va da, respectively.

    I conclude these issues with two further observations. First, since (Zwicky, 1977)with the Romanian examples quotes Hetzron (1976 ms.), which, in turn, quotes Olsen(1928:80), such errors propagates in other articles and can lead to distorted descriptionsof language phenomena. Second, the use of hyphen as element of the meta-language in theLeipzig Glossing Rules interferes with its use as orthographic element in object languages,e.g., in Romanian. Since the hyphen in Romanian is a frequently used character withalready overloaded functionality this fact adds more ambiguity to the already extant one.Is there a way out from this situation?

    6 SolutionsAs in philosophy, putting the appropriate questions is more valuable than trying to giveanswers, so in problem solving, making out the real problems can lead to clear solutions.Lets start to formulate the solution of the very last issue, the ambiguities added by theuse of common elements both in the meta-language and in the object language.

    16

  • Since the Leipzig Glossing Rules are meant to work for all languages a possible solutionof avoiding ambiguities between the two types of language is to collect all elements fromall languages with writing systems and exclude them from the meta-language. This wouldbe a tedious and, perhaps, not that viable solution because the set of common elementscould be large and, in the meta-language, one has to provide a replacement for eachelement belonging to both.

    The simplest solution has already been used in Fig. 4: the correct orthographic stringof the example in (Klein, 2007[p. 63, Ex.14]) has been added as the top line. Thatmeans, in order to avoid ambiguities in languages that have a standard orthography, eachexample has to contain an additional line: the language data in the correct spelling. Inthe spelling line no meta-language elements should be used, while in the analysis andglossing lines, all possibly ambiguous elements from the object language are banned.With other words, each set of ambiguous elements between the two type of languagesbe interpreted as belonging only to its specific line. Hyphens in the orthography stringshould be always hyphenobjectlanguage while hyphens in the analysis and glossing linesshould be exclusively hyphenmetalanguage. If needed, the time stamp of the orthographyused has to be provided in case of orthography changes over time.

    For my own examples in this article, although I have not used the Leipzig GlossingRules I added the correct orthography in the string in square brackets on the right ofeach example. Even if sometimes it seems redundant the LaTeX example style used hasto add whitespace characters in order to align the items in the first and second line theanalysis and the glossing lines , a fact that distorts the correct orthography.

    Lets turn now to the remaining ambiguities in the description of Romanian weak pro-nouns. Again, crystallizing the individual problems offers at the same time the solution:Disambiguate each type of ambiguity!

    To get a correct description of Romanian weak pronouns in terms of syllabicity, thereare three problems to pay attention for. First, since it has been shown that the -prothesis is not pronoun proper part all -prothetic forms be classified as non-syllabic.Second, in pre-verbal position the occurrence of a hyphen is unequivocal, it marks asandhi, i.e., desyllabification. Hence, the only problem here is to disambiguate betweensyllabic and non-syllabic forms with the same graphemic representation such as syllabicmi in mi-l /mil/ or mi-i /mitsj/ vs. non-syllabic mi in s-mi /s@mj/, mi-o /mjo/, ormi-a /mja/. Third, in addition to the disambiguation between syllabic and non-syllabicforms, the challenge in post-verbal position is to disambiguate between the hyphen usedas post-verbal marker, the hyphen used as sandhi marker, and the hyphen used as both

    17

  • syllabic non-syllabici-l dai, i le dai, d-i-le nu-i dai, nu i dai, s-i dai,

    i dai, i-ai dat, d-i

    le dai, d-i-le, d-le acum le-ai dat, d-le-acum

    te duci, du-te acas te-ai dus, du-te-ai, du-te-acasa

    se apuc, fac-se lumin, mi se face ru se-apuc, s-apuc,s-a apucat, fac-s-ar lumin

    Figure 6: Classifying Romanian weak pronouns i, le, te, and se into syllabic vs. non-syllabic

    post-verbal and sandhi marker.Table 1 shows some instances of disambiguation on all levels by means of the following

    marking:

    an oval box denotes a non-syllabic part of a sandhi (e.g., l )

    consequently, syllabicity is represented by the lack an oval box around an item(e.g., i , mi le faci, f mi le)

    a thick hyphen denotes solely a sandhi (e.g., i l )

    a half-thin/half-thick hyphen denotes both sandhi and post-verbality(e.g., cumpr

    l ) a thin hyphen denotes solely post-verbality (e.g., f mi le)

    The result of such cumbrous disambiguation work with the Romanian weak pronounsis a clean classification of the items under scrutiny into syllabic vs. non-syllabic, as isdone exemplarily in Fig. 6 for some instances of i, le, te, and se.

    7 ReflectionsLet me step back and see what I have done. I evaluated different models and descriptionsfor Romanian weak pronouns, I pointed to the weaknesses of the models and descriptions,I found out that the reasons for the errors are different types of ambiguities, and finally,I disambiguated all these types of ambiguities. And what is the result of all this work?

    18

  • No. Orthography Disambiguation Translation1. i-ai cumprat cartea.

    i ai cumprat cartea. Youve bought the book.2. i-l cumperi. i

    l cumperi. You buy it.3. i-o cumperi.

    i o cumperi. You buy it.4. S-i cumperi cartea! S

    i cumperi cartea! Buy the book!5. Te duci acas. Te duci acas. You go home.6. Du-te acas! Du te acas! Go home!7. Du-te-acas! Du

    te acas! Go home!8. Cumpr-l! Cumpr

    l ! Buy it!9. Cumpr-i-l! Cumpr i

    l ! Buy it!10. Cumpr-i cartea! Cumpr

    i cartea! Buy the book!11. Cumpr-i-o! Cumpr

    i o! Buy it!12. M lai n pace. M lai n pace. You let me in peace.13. Las-m n pace! Las m n pace! Let me in peace!14. Las-m-n pace! Las m n pace! Let me in peace!15. O lai n pace. O lai n pace. You let her in peace.16. Las-o n pace! La s o n pace! Let her in peace!17. Las-o-n pace! La s o n pace! Let her in peace!18. Mi le faci acum. Mi le faci acum. You do them for me now.19. F-mi-le acum! F mi le acum! Do them for me now!20. F-mi-le-acum! F mi

    le acum! Do them for me now!21. Ceva se aude. Ceva se aude. Something can be heard.22. Ceva se-aude. Ceva se aude. Something can be heard.23. Ceva s-aude. Ceva s aude. Something can be heard.24. Tu te auzi. Tu te auzi. You can be heard.25. Tu te-auzi. Tu

    te auzi You can be heard.26. Du-te ncolo! Du te ncolo! Go away!27. Du-te-ncolo! Du te n colo! Go away!28. Du-te n cas! Du te n cas! Go inside!29. Du-te-n cas! Du te n cas! Go inside!30. Te mbei. Te mbei. You get drunk.31. Te-mbei. Te m bei. You get drunk.Table 1: Hyphen and syllabicity disambiguation for pre- and post-verbal Romanian weak pronouns

    19

  • Well, the climax is a cheap partition of weak pronouns into syllabic and non-syllabic,something an eighth-grader in a Romanian school should have a good command of. Wasit worth it? Absolutely!

    The first insight is that descriptions of Romanian weak pronouns that do not try tomodel, hence, formalize them have a better coverage and are better suited to captureproperties that can be intersubjectively tested. This is the case with descriptions in(Iliescu, 1975) and (Bredemeier, 1976), as opposed to those in (Dobrovie-Sorin, 1999),(Monachesi, 2000),(Klein, 2007), and (Svescu Ciucivara, 2009), (Popescu, 2003), or(Chereche, 2014).

    The second insight is that descriptions on the morpho-syntactic and prosodic level,i.e., on the low-level, are much better suited than those trying to model the phenomenainto syntactic frameworks. This is the case with (Chereche, 2014), (Popescu, 2003), or(Gerlach, 2002), as opposed to (Dobrovie-Sorin, 1999), (Monachesi, 2000), (Klein, 2007),and (Svescu Ciucivara, 2009). On the other hand, low-level models have a narrow,sometimes perhaps too narrow, view on the input string. While syntactic approachesoperate at least on clause-level, (Popescu, 2003)s Optimality Theory model tries hardto cope with optional sandhi (cf. Ex. 32 vs. Ex. 33), and (Chereche, 2014)s model again, an Optimality Theory model ignores optional sandhi phenomena altogether bydeeming them dependent on many uncontrollable factors.

    The third insight is that there is little scientific communication between the high-level and low-level models, if at all. By scientific communication I mean comparingapproaches on a specific level of granularity or even trying to integrate the own approachin the other. Refining the old label clitics into syntactic clitics vs. phonologicalclitics as in (Dobrovie-Sorin, 1999), or into internal clitics vs. affixal clitics asin (Chereche, 2014) does not lead to a better understanding of the phenomena nor to abetter understanding of each others model.

    Further, (Klein, 2007)s categorization into strong pronouns, weak pronouns, andclitics is not totally congruent to (Cardinaletti and Starke, 1999)s partition into strongpronouns, weak pronouns, and clitics. And although is seems that Romanian isperfectly suited for such an tripartite analysis as in (Cardinaletti and Starke, 1999), thearticle does not take into account the Romanian pronominal system at all. This mighthave to do with their method, as it is stated that [w]hat is to be explained by themodel are (genuine) generalisations. In such research, it is usual that some facts resistgeneralisation, and some generalisation resist integration into the model. (Cardinalettiand Starke, 1999[p. 44]).

    20

  • Figure 7: U. Klein, Encoding of argument structure in Romanian and SiSwati, PhD thesis, Universityof London, 2007: p. 77

    Generalization is the goal of any science, just collecting and counting entities wouldrender a scientific domain into an accounts department. Yet, generalization implies aprocess starting with specific instances, hence it is a bottom-up process. If you startto generalize from a far to higher level you can never accommodate for the instancesyou ignored, these are the facts that resist generalisation, in (Cardinaletti and Starke,1999)s terms, the black swans of modus tollens3.

    Facts that resist your generalization are in fact counterexamples, facts that discardyour generalization, and in stating that, I am not alone with this opinion. An introduc-tory book in critical thinking expresses the same ideas as follows: When we generalize,we make a general claim about something based on specific evidence about that some-thing. [] In fact, those that are universal statements (claims about all or none, alwaysor never) are often overgeneralizations (an error []); all it takes to disprove such anovergeneralization is one counterexample. (Tittle, 2011[p. 287]).

    Generalization is not only a theoretical issue, a good generalization enables a goodmodeling of phenomena, hence a workable, operational, implementable model. At the be-ginning of this section I stated that the result of my hard disambiguation labor is a cheappartition of weak pronouns into syllabic and non-syllabic, something an eighth-grader ina Romanian school should know. Yet, it is obvious that the analysis of -prothetic formsas non-syllabic is an abstraction that has not been employed by any other descriptions,models, or formalizations. Ignoring this abstraction would lead to, at least, implausibleframeworks such as (Klein, 2007)s, where all possible -prothetic forms are modeled asinput items as such (see Fig. 7). This is tantamount to a model of the human world inwhich most babies are born with a prothesis, say, a wooden leg.

    The generalization over -prothetic and non- -prothetic is on a very low level, on thelevel of syllable. Can we work with weak pronouns or clitics without giving a definitionfor them? Well, there are two basic types of definitions: intensional and extensional.Some researchers came up to the conclusion that the term clitic is best thought of as

    3http://en.wikipedia.org/wiki/Falsifiability#Inductive_categorical_inference

    21

  • an umbrella term for a variety of properties which may or may not coincide. Although itmay be useful in individual languages to set up a special sub-category of clitic to capturevarious distributional or formal generalizations, there is no obvious sense in which cliticsrepresent a uniform, universal category. (Spencer and Lus, 2012[p. 11])

    The universe of whatever I refer to as Romanian weak pronouns is so small thatit is manageable even with an extensional definition, i.e., an exhaustive item listing.(Chereche, 2014) refers to them as pronominals (as opposed to stressable pronouns)without giving an intesional definition for them.

    The main message of the present article is that it is more important to describe theitems under scrutiny carefully, i.e., with both their phonologic and their writing repre-sentation, and as surface-oriented as possible, i.e., with as little theory-laden elementsas possible than to try to give intensional definitions for clitics or Romanian weakpronouns.

    But we have to keep in mind that, as mentioned, science results are communicatedalmost exclusively in written form. The written phonetic representation, e.g. the IPAcharacter set, is a more accurate mapping between phoneme and grapheme than the or-thography of a language but it could create even more understandability problems amongresearchers. The non-syllabic form of the Romanian weak pronoun mi in combinationwith an auxiliary verb as its syllabic support as in mi-a, for instance, is represented in(Iliescu, 1975) in a different way in Fig. 3 as in (Chereche, 2014) in Fig. 2.

    And even if such a partition into syllabic vs. non-syllabic I arrived at is not un-questionable because of the category syllable (cf., for instance, (Blevins, 1996)) is notunquestionable there is a far greater chance that more researchers would agree uponthe category syllable than upon categories such as (Dobrovie-Sorin, 1999)s syntacticclitic or (Chereche, 2014)s internal clitic or affixal clitic.

    Why is an agreement among researchers relevant? Speaking about natural sciences,(Popper, 2009) states that [] perceptions and observations (more precisely, perceptionand observation reports) are never taken seriously in science unless they areobjective, that is, intersubjectively testable (Popper, 2009[p. 131]).

    Is objectivity, i.e., intersubjective testability, possible in linguistics? By stating thatit is usual that some facts resist generalisation, (Cardinaletti and Starke, 1999) deny apriori any other researcher even the least chance to falsify their genuine generalisation.The only thing worse than a presumed impossibility of falsifiable hypotheses in linguisticsis a hostile attitude towards a sensible research argumentation and communication.

    22

  • 8 ConclusionsA prima facie simple question of the type What is a clitic? entails fundamental philo-sophical problems of different kinds. Initially an ontological question What is the entity? has necessarily to do with humans possibility of recognizing its features, hencean epistemic problem What can I recognize? which, in turn, leads to the questionof subjectivity of perception Do we humans recognize the entities features the sameway? Yet, to test that we recognize the entities features the same way we have tocommunicate our research results to each other.

    The next question is whether we understand each others terms, each others datadescriptions, each others models the same way. Is this the case? It is hard to say.We re-define old terms and invent new terms, we overload old terms with new meaningsand refine established terms by further sub-classifying them. I am not alone in statingthat: we often find that different experts working on different language groups will usethe same terms but with different meanings, or alternatively will use different terms todescribe essentially the same phenomenon. (Spencer and Lus, 2012[p. 6])

    One can think that it is not possible to formulate falsifiable hypotheses in linguisticsbecause of the huge amount of assumptions or because of gradiency of linguistic categories.One can interpret Poppers falsifiability of hypotheses as prescriptivism and Kuhnsparadigm shift as descriptivism (Tore Nesset, personal communication), yet I do notshare this view totally. In my humble opinion, Poppers falsifiability method is applicableon how each individual corrects and adapts his world knowledge as a function of experienceaccumulation. To exemplify this idea, let me tell you a joke:

    A woman goes to Court in order to sue her son-in-law for defamation.Woman: My son-in-law called me a hippopotamus.Judge: When did that happen?Woman: About 25 years ago?Judge: Why have you waited so long with the complaint?Woman: It was not before yesterday that I went to the zoo and saw one.

    Since the woman thought a hippopotamus was some term of endearment she didnt havea reason to sue her son-in-law for insulting her. The encounter with a hippopotamus atthe zoo was the falsification of her believes, the falsification of her hypothesis about themeaning of the word hippopotamus.

    Ergo, in our communication, be it in private or in research, we will never know whetherwe understand each other the same way, we might only find out that in a specific point

    23

  • our apprehensions were not congruent with each other. Or, to put it in Poppers terms:we can not prove that our apprehensions are congruent with each other, we can onlyfalsify it. Call it solipsism if you want, I call it sound skeptical realism, anyway, wewill never know whether we understand these terms the same way.

    ReferencesBarbu, A. M. (1999). Complexul verbal. In: Studii i Cercetri Lingvistice, pp. 3984.Blevins, Juliette (1996). The Syllable in Phonological Theory. In: The Handbook of

    Phonological Theory. Ed. by John A. Goldsmith. Blackwell.Bredemeier, Jrgen (1976). Strukturbeschrnkungen im Rumnischen. Studien zur Syntax

    der pr- und postverbalen Pronomina. Tbingen: TBL Verlag Gunter Narr.Calude, Andreea S. (2001). Romanian clitics: Siding with the Serbo-Croatian or the

    French? In: Revue roumaine de linguistique 46.1-4, pp. 91104.Cardinaletti, Anna and Michal Starke (1999). The typology of structural deficiency: A

    case study of the three classes of pronouns. In: Clitics in the Languages of Europe.Ed. by Henk C. van Riemsdijk. Mouton de Gruyter.

    Chereche, Anca (2014). A Prosodic Analysis of Romanian Pronominal Clitics. In: Uni-versity of Pennsylvania Working Papers in Linguistics. Vol. 20. 1.

    Dindelegan, Gabriela Pan, ed. (2013). The Grammar of Romanian. Oxford UniversityPress.

    Dobrovie-Sorin, Carmen (1999). Clitics across categories: The case of Romanian. In:Clitics in the Languages of Europe. Mouton de Gruyter.

    Engh, Jan (2006). Norwegian examples in international linguistics literature: An inventoryof defective documentation. UBO Skrifter 32. Universitetsbiblioteket i Oslo.

    Gerlach, Birgit (2002). Clitics Between Syntax and Lexicon. John Benjamins.Guu-Romalo, Valeria, ed. (2008). Gramatica limbii romne. Editura Academiei Romne.Harris, Alice C. (2002). Endoclitics and the Origins of Udi Morphosyntax. Oxford: Oxford

    University Press.Iliescu, Maria (1975). Pentru o sistematizare a predrii pronumelui personal neaccentuat

    romnesc (la studenii strini). In: Limba Romn 24, pp. 5162.Klavans, Judith L. (1995). On Clitics and Cliticization: The Interaction of Morphology,

    Phonology, and Syntax. New York: Garland Pub.Klein, Udo-Michael (2007). Encoding of argument structure in Romanian and SiSwati.

    PhD thesis. University of London.

    24

  • Legendre, Graldine (1999). On the status and positioning of verbal clitics. ms., JohnsHopkins University.

    (2000). Positioning Romanian Verbal Clitics at PF: An Optimality-theoretic Anal-ysis. In: Clitics from Different Perspectives. Ed. by B. Gerlach and J. Grijzenhout,pp. 219254.

    Lombard, Alf (1976). Le prosthtique du roumain. In: Acta Societatis linguisticaeUpsaliensis. Vol. 2. 5.

    Monachesi, Paola (2000). Clitic placement in the Romanian verbal complex. In: Cli-tics in phonology, morphology, and syntax. Ed. by B. Gerlach and J. Grijzenhout.Linguistik Aktuell. Amsterdam: John Benjamins.

    (2005). The Verbal Complex in Romance: A Case Study in Grammatical Interfaces.Oxford University Press.

    Popescu, Alexandra (2000). The morphophonology of the Romanian clitic sequence. In:Lingua. Vol. 110, pp. 773799.

    (2003). Morphophonologische Phnomene des Rumnischen. PhD thesis. Universityof Dsseldorf.

    Popper, Karl Raimund (2009). The Two Fundamental Problems of the Theory of Knowl-edge. Routledge.

    Soare, Gabriela (2005). Romanian Syntax for FiPS. Tech. rep. Laboratoire danalyse etde Traitement du Langage (LATL), University of Geneva.

    Spencer, Andrew and Ana Lus (2012). Clitics: An Introduction. Cambridge UniversityPress.

    (2013). Canonical Morphology and Syntax. In: ed. by Dunstan Brown, Marina Chu-makina, and Greville G. Corbett. Clarendon Press. Chap. The canonical clitic.

    Svescu Ciucivara, Oana (2009). A Syntactic Analysis of Pronominal Clitic Clusters inRomance - The view from Romanian. PhD thesis. New York University.

    Tittle, Peg (2011). Critical Thinking: An Appeal to Reason. Routledge.Yngve, Victor H. and Zdzisaw Wasik, eds. (2004). Hard-Science Linguistics. London:

    Continuum.Zwicky, Arnold (1977). On clitics. In: Bloomington: Indiana University Linguistics Club.Zwicky, Arnold M. and Geoffrey K. Pullum (1983). Cliticization vs. inflection: English

    nt. In: Language 59, pp. 502513.

    25

    IntroductionClitic-host pairTo be or not to be a clitic or an affix or anRelevant featuresAmbiguities in descriptionSolutionsReflectionsConclusions