The role of the attraction in agreement mistakes in English vs Spanish

Facultade de Filolox´ıa Grao en Lingua e Literatura Inglesas TRABALLO DE FIN DE GRAO. Curso: 2018-19. Xullo de 2019 The role of the attraction in agreement mistakes in English vs Spanish AUTORA: Mar´ ıa Su´ arez ´ Alvarez TITOR: Carlos Acu˜ na Fari˜ na Facultade de Filolox´ıa Grao en Lingua e Literatura Inglesas TRABALLO DE FIN DE GRAO. Curso: 2018-19. Xullo de 2019 The role of the attraction in agreement mistakes in English vs Spanish AUTORA: Mar´ ıa Su´ arez ´ Alvarez TITOR: Carlos Acu˜ na Fari˜ na SUAREZ ALVAREZ MARIA - 3271152 0T Firmado digitalmente por SUAREZ ALVAREZ MARIA - 32711520T Nombre de reconocimiento (DN): c=ES, serialNumber=IDCES-327 11520T, givenName=MARIA, sn=SUAREZ ALVAREZ, cn=SUAREZ ALVAREZ MARIA - 32711520T Fecha: 2019.06.21 19:20:30 +02'00' Contents Abstract ...................................................... 7 List of key words ................................................. 9 1 Introduction .................................................. 11 2 Agreement in Linguistics .......................................... 17 2.1 Mainfeaturesofagreement....................................... 18 2.1.1 Gender .............................................. 18 2.1.2 Number.............................................. 19 2.1.3 Person .............................................. 20 2.1.4 Case................................................ 20 2.1.5 Other main features of agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.2 Linguistic theories of agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.2.1 Formalviewofagreement.................................... 21 2.2.2 Unification............................................. 22 2.2.3 MixedAgreement ........................................ 23 2.2.4 AgreementHierarchy ...................................... 25 3 Agreement in psycholinguistics ...................................... 29 3.1 Experimental work from the 90’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.1.1 Bock&Miller(1991)....................................... 31 3.1.2 Vigliocco et al (1995; 1996) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.1.3 Eberhard(1997) ......................................... 34 3.2 Main models of agreement in psycholinguistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3.2.1 MarkingandMorphing...................................... 35 3.2.2 MaximalInput........................................... 38 3.3 Conclusion ................................................ 41 v 4 My study .................................................... 43 4.1 Semanticforces:Concreteness..................................... 43 4.1.1 Previous studies on concreteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4.2 Mystudy ................................................. 47 4.2.1 First study: concreteness test (Spanish) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.2.1.1 Description....................................... 48 4.2.1.2 Objective........................................ 48 4.2.1.3 Method......................................... 48 4.2.1.3.1 Participants.................................. 48 4.2.1.3.2 Materials................................... 48 4.2.1.3.3 Procedure .................................. 49 4.2.1.4 Results......................................... 49 4.2.1.5 Discussion ....................................... 49 4.2.2 Second study: concreteness test (English) . . . . . . . . . . . . . . . . . . . . . . . . . . 50 4.2.2.1 Description....................................... 50 4.2.2.2 Objective........................................ 50 4.2.2.3 Method......................................... 50 4.2.2.3.1 Participants.................................. 50 4.2.2.3.2 Materials................................... 50 4.2.2.3.3 Procedure .................................. 51 4.2.2.4 Results......................................... 51 4.2.2.4.1 Parallelresults................................ 51 4.2.2.5 Discussion ....................................... 52 5 Conclusion and suggestions for further research ............................ 53 A Appendix: Questionnaire used in the concreteness study in Spanish ................ 57 B Appendix: Questionnaire used in the concreteness study in English ................ 59 References ..................................................... 60 vi 1 \t ll l 1 :\UI LH 111 Ol ():\ 1,\ CU BR IR ES TE F ORMULARIO ELECTRONICAMENTE UNIVERSIDADE DE SANTIAGO DE COMPOSTElA FACULTADE DE FILOLOXfA r-;~·OUT. 2018 ENTRADA Formulario de de imü:-a d'll~e:..tit u~ o.:e:. r e sum Traballo de Fin de Grao curso 2 018/ 2019 APELIDOS E NOME: Suárez Álvarez, Marí a GR AO EN: Lingua e Literatura Inglesas (NO CASO DE MODERNAS) MENCIÓN EN: TITOR /A: Ju an Carlos Acuña Fariña LI ÑA TEMÁTI CA ASIGNADA: Procesamento sintáctico e s in taxe en lingu a inglesa SO LICITO a aprobación do seg uint e título e resumo: Tftulo: The role of the attraction in agreement mistakes in English vs Spanish. Resumo [na lingua en que se vai redactar o TFG ; e ntre 1000 e 2000 ca racter es ]: Attraction is a linguistic phenomenon that occurs when a competition between two or more NPs in si de of a larger NP cause mistakes in th e ag reement of rela tion ships, li ke subj ec tverb, of the kind: '*the label on the bottl es are broken'. In the English Langu age it is estimated that 13% of compl ex NP s establi sh inco rr ect ag r ee ment with the verb. The main object of this study is to analy se agreement (or proximi ty concord) and mistakes relat ed with it due to att r ac tion from an experimental point of view. Th eore ti ca ll y, agreement can be con ce ived fr om two different points of view in linguistics: th e formalist and the functional 1 cognitivist, depending on whether th ey take syntax as se parat ed from cognition oras a reflection of semanti cs. T hi s work will be organised in th e following way: a th eor et ical background about these two visions will be presented, as well as a discussion of agreement from both a linguistic and a psycholinguistic po int of view; the n my own re sea rch on attraction interferen ce in agreement will follow. This research will be based on completive questionnair es answered by Eng lish and Spani sh speakers in order to s tudy the pre se n ce of attraction in agreement mistakes in both languages and to be able to make a comparison between the mistakes in th ese languages. Finally, 1 will summari se my main findin gs and provide a gener al discussion of the topi c in the light of the se. SRA. DECANA DA FACULTADE DE FI LOL OXfA (Pr es ide nt a da Co misión de Tí tul os de G ra o) Santiago de Compo stel a, 31 de Outubr o de 20JK_. Sinatura do/a int eresado/a Vi sto e prace { si natura do/a titor/ a) 1 6 NOV . 2018 --~-_,. ) de F ilol oxía SRA. DECANA DA FACU LTADE DE FI LOLO XfA {Pr es id e nta da Co mi sión de Títulos de Grao) LIST OF KEY WORDS agreement,agreement mistake,attraction,feature. 9 16 such as Italian and Spanish, where she found distributivity effects, thus, semantic interference in agreement processes. When she tried to find these same effects on English, her studies reflected the data provided by Bock & Miller (1991), which led authors to the conclusion that languages have different sensitivities to distributivity forces. Later on, Eberhard (1997) proved that if the preambles were correctly manipulated, every language showed effects of distributivity in their agreement processes. These findings and perspectives are reflected in two main models: the hybrid model of Marking and Morphing, presented by Bock, Eberhard, Cutting et al in 2001, where agreement is considered an encapsulated process divided into phases; and Vigliocco’s (2001) functionalist model, Maximal Input, with an opposite perspective to the one formulated in Marking and Morphing, that is, that agreement phases can affect each other if convenient for the creation of an interpretation. After introducing the theoretical framework on which I have based my study, I introduce my own experiment that has the aim of finding some evidence of semantic interference on agreement. Since it is inspired on the experiment conducted by Riveiro Outerial (2014) and she did find some evidence that points towards the presence of semantic interference, we expect to find something similar. Apart from testing agreement mistakes we also explore the semantic constraint of concreteness, that is, whether there is any difference on the agreement mistakes when the head noun is abstract or concrete, as abstract nouns are more difficult to process, this advantage presented by concrete nouns should be present in the process of agreement. The study of the semantic constraint of concreteness is complex due to the fact that making a differentiation between concrete and abstract nouns is not as clear as we could think, words like music, which are traditionally considered abstract can, on the other hand, be perceived by senses such as the hearing. However, concreteness has been studied by some authors who presented different theories that intend to explain these differences regarding the processing of abstract and concrete words. These theories are The dual coding theory (Pavio, 1971, 1991) and The context availability hypothesis (Bransford & McCarrel, 1974; Schwanenflugel & Shoben, 1983). Both of them propose different reasons for the differences in the processing of the two types of words, but they agree on the existence of this difference. After presenting some information on what concreteness is and previous studies on the topic, I continue with my study, which will try to test the hypothesis that concrete nouns are more easily processed than abstract nouns and, therefore, will produce less agreement mistakes. In order to do this, we conduct an experiment similar to the one conducted by Riveiro Outeiral (2014), that is, following the same method used by Bock & Miller (1991) consisting in completion tests, but making some corrections in order to avoid problems that Bock & Miller faced when analysing their results. I proceed now to discuss agreement and attraction, beginning with agreement in linguistics. 2AGREEMENT IN LINGUISTICS As we have previously pointed out in the introduction, agreement consists in a rule according to which a singular controller requires a target with singular features, and a plural controller requires a target with plural features. This definition, together with the fact that agreement is more or less present in more than half of the known languages of the world, makes it seem like something simple, common and even straightforward. Nevertheless, I will now introduce some varieties that present cases of agreement that we could consider “exotic” and will prove that agreement is a more fascinating phenomenon than it may seem. For instance, the Bengali language, spoken in Southern Asia, presents a rather complete and complex system: Bengali’s pronouns have forms for 12 different categories of person, number and aspect, and its verbs have syncretized tense/mode/aspect and person suffixes with widespread homophony, but for each tense/mode/aspect set of forms, there is independent marking of five person/number/respect classes. The number of combinations is explosive. For instance, -i is the first person ending in the present, but the second person inferior in the future, while –o is the second person ordinary in the present and first person in the future. (Acu˜ na Fari˜ na 2009:390) Another variety that presents an “exotic” agreement system is Basque, or Euskera, the language spoken in the Basque Country and in some areas in the South West of France. This language, unlike Spanish or English is an ergative language (vs active language). This means that in this type of language, the only participant of an intransitive construction (normally what we, in Spanish or English, would call a subject) has the semantic role of patient, while in Spanish or English would have the role of agent. Regarding Basque from the point of view of agreement, which is the interesting aspect for the present piece of work, it stands out the fact that in this language the verb has to agree, not only with the subject, as it happens in English and Spanish, but also with the Direct and Indirect Object: Aitak librua ekarri dio semeari (‘The father brought the book to the son’); Aitak libruak ekarri dizkio semeari (‘The father brought the books to the son’); Aitak libruak ekarri dizkie semeei (‘The father brought the books to the sons’) (Villasante Kortabitarte 1978:12). 17 18 2.1 Main features of agreement 2.1 Main features of agreement Despite the different ways in which languages show agreement, some of the most traditional literature that deals with this topic has called our attention to the fact that there are some elements that are present in every agreement operation. These elements are controllers,targets,features and domains. After the introduction of the other terms in the previous chapter, I will now pay especial attention to the different features of agreement, which have the function of indicating in what respect there is agreement. There are features that are present in languages in a more frequent way, like number and gender, but these are not the only features affected by agreement, there are others such as person,case,aspect,mood,polarity, etc. I will now present each of them separately and provide examples. 2.1.1 Gender Starting with gender, which is a feature that is “mainly reflected in the adjectives that acquire the noun’s gender marks” (Riveiro Outeiral 2014:29). First of all, in order to understand the different systems of gender agreement present in the word’s languages, we should distinguish between biological (or natural) and arbitrary (or grammatical) gender. In his Dictionary of Linguistics and Phonetics, David Crystal (in Crystal 2008:206) defined biological gender as the grammatical category “where items refer to the sex of real-world entities”, and grammatical gender as something “which has nothing to do with sex, but which has an important role in signalling grammatical relationships between words in a sentence”. Now that we have made this distinction clear, we can move on with the explanation of this feature. Since the value of the arbitrary gender has to be available in a language’s lexicon, it is a lexical feature for the majority of the languages. Speakers have to learn the gender of every noun in order to make the appropriate agreement ties. Depending on the parameters used to assign nouns to a particular gender class, there are different systems of agreement: semantic and formal agreement. In the case of semantic agreement, “nouns are assigned to a gender class following a purely semantic criterion” (Riveiro Outeiral 2014:29). In other words: nouns’ gender depend on the sex of the notion that this noun represents. Male entities will present a masculine gender (gato), female entities will present a feminine gender (gata) and all other words will be assigned to a neutral category (pez). In the case of formal agreement systems, “the noun’s form is responsible for gender” (Riveiro Outeiral 2014:30). Since the majority of the words do not have biological sex (mesa), the first system is not enough to assign these words to a gender class, so they fall into a phonological or morphological classification. This is the case of Spanish, in which words ending in –o are usually masculine, and words ending in –a are normally feminine, as previously illustrated with the gato/a example. However, this is not the case in every word, as there are examples such 2 Agreement in Linguistics 19 as mano (feminine) and mapa (masculine), or words that do not end in –o or –a, such as an´ alisis (masculine) or pared (feminine). Summarizing, the allocation of a word to a particular gender class can follow either of the criteria or a combination of both. 2.1.2 Number Number is a feature strongly based on meaning. Normally, the singular denotes one entity while the plural denotes more than one. In this sense, number is difficult to conceptualise in an arbitrary fashion, like gender. This does not mean that number is something monolithic crosslinguistically, since not all languages present a distinction of singular vs plural (cat vs cats). Some languages have other values such as dual for two entities, present in Irish; trial for three, as in Larike; or paucal for a small number of entities, as in Bayso (Corbett 2006:130). Other languages might not have any number marking at all. Even in languages that present what for us is the most common system (only singular and plural) there are some special cases like nouns that are only present in either singular or plural form, the so called singulare tantum (love) and plurale tantum (pants). The usual way of marking number in many languages is through the addition of a morphological ending (in Spanish and English a –s: mesa vs mesas or table vs tables). However, this is not the only way of marking the plural. Some languages, like Russian, mark the plural form through prosodic differences, depending on the part of the word on which stress is placed; other languages mark number through alternations that depend on the morphological environment, like the case of Macedonian1; other varieties may use augments in the singular, plural or both, like the case of some words in Russian2; languages like Ilocano, variety spoken in some parts of the Philippines use reduplication, that is, the repetition of some part of the lexical item3; other languages present internal modifications of the plural forms with respect to the singular, like the case of goose in English (geese in plural); other varieties remove material instead of adding it to the word, as is the case of Hessian German4. As we have seen, the systems to mark plural forms presented in languages go from adding phonemes to the stem to more complex strategies. 1E.g. koren (root) plural koreni (roots). However, other endings may imply a change in some of the letters. 2E.g. bolgar-in (Bulgarian) plural bolgar-y (Bulgarians). 3E.g. p´ usa (cat) plural pus-p´ usa (cats). 4E.g. hond (dog) plural hon (dogs). 20 2.1 Main features of agreement 2.1.3 Person Another frequent feature of agreement in the world’s languages is person. “Person makes reference to participants in the speech act” (Riveiro Outeiral 2014:36). This feature is typically present in pronouns and verbs. For pronouns person is an inherent feature, meaning that it is an inherent part of its meaning; while for verbs it is contextual, that is, they acquire person by agreement. Languages tend to have three persons for the singular and three persons for the plural -first person makes reference to the speaker, second to the addressee(s) and third to people different from the speaker and the addressee-. However, some languages make a subdivision in any of its persons, splitting, as is the case of some Algonquian languages, the third person into more and less central participants in the situation. 2.1.4 Case Case is not considered by Corbett (2006) a canonical feature of agreement. This is due to the fact that case “does not reflect an asymmetric relation similar to the case of gender or number” (Riveiro Outeiral 2014:36). While in gender and number there are controllers with certain features that are expressed in targets, case depends on the syntactic environment and therefore it is not an inherent feature of the controller. 2.1.5 Other main features of agreement Finally, tense,aspect,mood and polarity present, with respect to the verb, a similar example to that of case with respect to the noun. They are usually features of the verb, “although in some languages other elements within the phrase may be marked for them as well, thus establishing a symmetric relation between elements” (Riveiro Outeiral 2014:37). As this symmetry goes against the nature of agreement as posited by formalist accounts, case, tense, aspect and polarity present some trouble to be considered canonical features of agreement. Corbett (2006) makes further distinctions between features and adds some more, such as definiteness and respect, however, they are more problematic than and not as common as the ones mentioned above, so I will not be commenting on them. 2 Agreement in Linguistics 21 2.2 Linguistic theories of agreement Now that the main features of agreement have been explained I will move on to different linguistic theories that contemplate agreement from different points of view. These models of grammar can be divided into those that view agreement as an “essentially encapsulated phenomenon whose range of operation is limited to the first cycles of the building of a clause structure” (Acu˜ na Fari˜ na 2009:390), and other models which maintain that agreement is “either penetrated by semantic forces or, more radically, a semantic process in nature” (Acu˜ na Fari˜ na 2009:390). 2.2.1 Formal view of agreement The first point of view that I will present is the formal view of agreement, which proposes a division into several encapsulated phases, phases that are psychologically unsupported, at least concerning agreement production. According to (Chomsky 1999:9) “phases are impenetrable [in order] to constrain computational load”. Phonological and semantic operations start once a syntactic phase has been completed, thus, the previous syntactic computations are deleted, becoming no longer available. These models assume that there is no communication between phases at any stage of the computation. These phases, or formal operations, that generation is divided into are: Merge,Agree and Move. Merge consists in the “fundamental structure-building operation” (Franck et al. 2006:6). It forms a minimal phrase out of two elements by stringing them together. If Merge operations are successively applied, it can “assemble the thematic nucleus of the sentence”- that is, the verb and its argumentsand further applications introduce the “functional structure of the sentence” (Franck et al. 2006:6) -that is, specifications of tense, aspect, mood, etc. Therefore, by these operations a configurational skeleton is created. It can be additionally modified by further applications of Merge and by another formal operation: Move. This allows speakers to displace elements that have already been introduced in the structure. This proposal implies that, unlike models which tend to assume “a single hierarchical representation over which all syntactic operations are computed [. . . ] several intermediate representations are postulated that reflect the cyclic derivation of the structure” (Franck et al. 2006:6). The other phase of agreement, Agree, of particular interest here, works as follows: The operation Agree is posited alongside Merge as a central component in the creation of sentential messages. Agree works on the premises of isolability, activeness and locality: once the thematic nucleus of the message is assembled, a functional structure AgrS is merged next. It is at that point in the derivation that AgrS enters in an Agree relation with the VP internal subject NP. The motivation for this is that AgrS has person 22 2.2 Linguistic theories of agreement and number features that need to be valued, so these are copied from the subject NP. The NP is turned active by possessing uninterpretable features that need to be checked inside the c-command domain of its goal. (Acu˜ na Fari˜ na et al. 2014:112) Several generative theoreticians have turned to attraction in order to test this type of theory. This is due to the fact that if attraction occurs as a result of the interference of semantic variables, such as underlying distributivity, that come from out of the hypothesized phase, the very existence of phases is called into question, or at least, in need of revision5. 2.2.2 Unification The second point of view of agreement that will be discussed here is the one presented by authors such as Pollard & Sag (1994) and Barlow (1992): unification. Unification theories do not contemplate agreement as a process in which controllers are present where agreement originates and targets receive the agreement features from the controllers, but instead, they see agreement as a matter of “cumulating partial information from the controller and the target” (Acu˜ na Fari˜ na 2009:394), information that is generated –even for the targets, in situ. According to these theories, the features of agreement originate where they are “actually spelled out” and carry “potential semantic import” (Acu˜ na Fari˜ na et al. 2014:111). These features are scattered all over the sentence, thus, this constant reduplication of form entails constant access to the meaning. Due to this scattering and reduplication, agreement is seen as a sort of ‘long component’ or a ‘discontinuous morpheme’. This also implies that features are simply shared instead of moved. “In these theories features are not slaves to the seriality of an initial syntactic cycle via strict step-by-step percolation, but can directly connect with the conceptual properties of the message at any time, [so] any semantic effect can be captured straightforwardly” (Acu˜ na Fari˜ na et al. 2014:111). This conception of agreement as a semantic phenomenon is a result of the many cases of agreement ad sensum6present in the world’s languages, especially in English. Normally, the number feature of the verb agrees with the number of the mentally identified referent of the subject. However, there are cases in which identifying the number of the subject is not so simple. Sometimes languages present nouns that are semantically ambiguous in number, like for instance, collectives, as they can denote the collective as a whole 5See chapter 3: ”Agreement in Psycholinguistics” for further information. 6The differences between types of agreement made by Corbett will be explained later on. However, I think it is important to clarify what we mean in this case by ‘agreement ad sensum’. Agreement ad sensum, or semantical agreement, is agreement consistent with its meaning. 2 Agreement in Linguistics 23 or as the individuals it comprises, therefore, being treated as singulars or as plurals: the team plays good vs the team play good. There are other cases of complex NPs functioning as subject, such as the picture on the postcards, in which there is a picture in each of the postcards. Although it may seem similar to the bridge to the islands, semantically it is not. The referent in the first case is only abstractly singular, but concretely “it encompasses the identical picture of multiple postcards” (Bock et al. 1991:51), however, in the second case, there is only one bridge to the islands. Thus, we might expect more “errors” in the sentences similar to the picture example than in the bridge example. Another example of this agreement ad sensum is found in number transparent nouns (like series,number,lot, etc.) that tend to establish this type of agreement in English, in which the agreement of the phrase is controlled by the ‘wrong’ controller, in this type of complex noun phrases in which the first element functions as a quantifier agreement occurs between the demonstrative and the semantically significant noun, e.g. a(sg) series of reports were (pl.) issued. This type of case of agreement in which the semantics of the noun overcomes formal agreement led authors to believe that agreement was a semantic operation. The main idea proposed by unification-based models is that, with rich morphology, “the two constituents that participate in an agreement relation may specify only partial information about a single linguistic object” (Acu˜ na Fari˜ na 2012:10). Thus, unification takes place when compatible information regarding the features on two sites becomes merged, and as we have said, features are shared, not copied or moved. Languages with rich morphology, such as Spanish and Italian, present verbs which contain this type of partial information directly and “can therefore directly connect to conceptual structure without any kind of mediation or control” (Acu˜ na Fari˜ na 2012:11). However, there are some strong arguments against treating unification as the only, or the major, cognitive strategy supporting agreement. One of these arguments is that if this were the case, there would be agreement inside adjective, adverbial and prepositional phrases as much inside NPs. Moreover, “unification per se does not relate to the Agreement Hierarchy [theory that will be explained later on], one of the most solid generalizations governing agreements system crosslinguistically” (Acu˜ na Fari˜ na 2018:16). 2.2.3 Mixed Agreement The third theory that I will present in this piece of work is proposed by Wechsler and Zlatic (2000; 2011) through a mixed model of agreement. First of all, in order to understand how mixed agreement occurs, we have to explain what these authors consider to be the relevant lexical features of the noun. These features are (1) semantic conditions on reference; (2) person, number and gender features of the referential index; (3) concord features; and (4) declension class. These four features are correlated by a chain of binary constraints. When individual constraints are violated, and thus, the chain is broken, this leads to the aberrant result of “intri- 24 2.2 Linguistic theories of agreement cate patterns of mixed agreement” (Wechsler et al. 2000:799). Cases in which the chain is broken and mixed agreement appears are, for instance, cases in which “a noun triggers one set of agreement features on adjectives but the NP headed by that noun triggers a different set of agreement features on verbs or coreferential pronouns” (Wechsler et al. 2000:799), that is the same controller imposing different feature specifications on different targets. The noun majestad in Spanish is an example of this phenomenon: in Su Majestad suprema est´ a contento. ´ El ha vuelto a casa., the word majestad refers to a male, thus, it triggers masculine agreement on predicates (contento) and pronouns ( ´ El), however, it triggers feminine agreement on attributive modifiers (suprema). As this example shows, this phenomenon can also happen with the gender features, it is not restricted only to number. The study of these cases is interesting for us, as they shed light on regular agreement processes. The data received from the study of cases that present mixed agreement led Wechsler and Zlatic to suppose that inflected nouns present two different feature sets that determine the agreement value it triggers: Concord features, which are closely related to the noun’s declension class, or Referential Index features, which are related to the noun’s semantics, regarding for instance if the noun denotes a male or a female. Concord and Index features are systematically related to each other. These features are normally identical, although in some exceptional cases in some languages they differ, as in the case of the Croatian words zena ‘woman’ and knjiga ‘book’, which are considered feminine nouns triggering feminine concord, but regarding the index-semantics correlations, while zena is female-denoting, knjiga is neither female nor male-denoting, so it “vacuously satisfies the rule” (Wechsler et al. 2000:800). That is, the biological and formal gender of zena (woman) are the same: a female entity which triggers feminine agreement. Knjiga (book), on the other hand, presents a conventionally adjudicated feminine formal gender, but the entity itself presents neither feminine nor masculine biological gender, thus, this correlation is not complete. However, this is not the only division that Wechsler and Zlatic (2000) propose. Index and Concord features are included into one of the two types of agreement that Wechsler and Zlatic distinguish: Grammatical Agreement. They argue that agreement processes can be divided into Grammatical Agreement and Pragmatic Agreement. While Grammatical Agreement is a result from structural properties of the grammar, Pragmatic Agreement is considered a result from “structural properties of the world as described in an utterance” and from “the general condition that coreferential elements must have compatible referential properties” (Wechsler et al. 2000:801). Then, Grammatical Agreement is divided into the Index and Concord features that have been mentioned above. So, summarizing: Wechsler and Zlatic (2000) divide agreement into Pragmatic and Grammatical Agreement, which is at the same time divided into Index and Concord features. These features act in different domains: since Index is attached to the referential index, it applies to NPs and pronouns, which are “referentially anchored” (Wechsler et al. 2000:807), and often they also refer to finite elements, like verbs and auxiliaries; but NP-internal elements that are not referential and lack referential indices, like determiners or adjectives, are 2 Agreement in Linguistics 25 referred to as concord. Normally, the gender/number features of Concord and Index for a given noun match, giving the illusion that “a single feature bundle on the noun is responsible for all the agreeing items: determiner, adjective, verbs and pronouns” (Wechsler et al. 2000:806). However, this is not always the case, and when these features do not match, mixed agreement takes place. 2.2.4 Agreement Hierarchy Finally, the contribution of Corbett to the Grammar of Agreement deserves special mention. Corbett introduced in 1979 his Agreement Hierarchy so as to explain when and how grammar would decide between semantic or syntactic agreement in a case where the formal and semantic properties of an item do not coincide, for instance: team is a collective noun that is grammatically singular, but it implies a plural meaning, a group of people, therefore, it is notionally plural. Nevertheless, in order to understand this proposal, we should know what he means when he makes a differentiation between semantic and syntactic agreement. For Corbett 2006:155), syntactic agreement (also called agreement ad formam, ‘formal agreement’ or ‘grammatical agreement’) is agreement consistent with the form of the controller. Semantic agreement (‘agreement ad sensum’, ‘notional agreement’ or ‘logical agreement’) is agreement consistent with its meaning. These terms are used only when there is potential choice. This committee is would be an example of syntactic agreement, as there is a singular noun agreeing with a singular determiner; while this committee are is an example of semantic agreement, as a committee is formed by several people, so it implies notional plurality and it agrees with a plural determiner. However, semantic agreement is restricted by the syntactic environment: *these committee sat late. In this case, committee does not accept plural agreement with the determiner. This proves that the effect that semantics has on agreement depends on the domain or syntactic environment in which agreement is taking place. As we have seen: the noun committee accepts either a plural or a singular verbal form, but only a singular determiner (Corbett 1979:203). Normally, formal and semantic properties of the controller coincide, thus, agreement is both syntactically and semantically justified, as in John has chosen sugar vs The Jones have chosen sugar. However, this is not always the case, for occasions in which the two types of agreement do not coincide, as with collectives, this hierarchy allows us to make predictions regarding the possibility and relative frequency of semantic agreement as compared to syntactic agreement. Good evidence to support this hierarchy is found in conjoined inanimate nouns in English, where the attributive must be singular, but the predicate and the relative pronouns can be of either number: This (*these) wind and rain have/has affected the harvest. When the conjoined nouns are animate, there is a shift in favour of semantic agreement: This (*these) man and woman were (*was) squatting in a castle. “Attributive position still requires singular agreement, however, the predicate requires plural agreement. The relative and personal pronouns require semantic (plural) 32 3.1 Experimental work from the 90’s work made by Bock & Eberhard (1993). Among these new findings we can include that (7) there was no difference between regular (boys) and irregular (men) plurals regarding attraction, and (8) while collectives like army and fleet did not result in any attraction, ordinarily inflected nouns like soldiers or ships did. These results facilitated links with the linguistics literatures on notions like directionality,feature inheritance, and/or feature copying and feature percolation, some of which will be mentioned again in this chapter. This inflectional account of agreement became also a copying account whereby “a controller which possesses inherent features passes them on to a target to establish an agreement relation. This is done in a formally encapsulated manner that is strongly reminiscent of cyclic phases in linguistics” (Acu˜ na Fari˜ na 2012:10). However, the conclusions reached regarding the null effects of semantic manipulations would have to be modified soon. 3.1.2 Vigliocco et al (1995; 1996) The next author that worked on this field and found out important results was Vigliocco. Bock & Miller (1991) had investigated the English language and concluded that distributivity had no effect on attraction mistakes. Vigliocco (1996) conducted research in Italian and found, for the first time, that in this language attraction patterns were sensitive to the distributivity of the preambles. Then, in 1996, Vigliocco et al. compared English and Spanish, in order to test if the distributivity effect previously found in Italian could be generalized to another language with a structure similar to Italian. Their premise was that if the hypothesis suggesting that “agreement is generally computed through a feature-copying operation” (Vigliocco et al. 1996:273) was right, these experiments should not reflect any difference between Spanish and English regarding distributivity effects; but if they did find any distributivity effect in Spanish, this would mean that, “at least in these languages, agreement is computed through a unification operation” (Vigliocco et al. 1996:273). This study led to more results that indicated that Spanish behaved like Italian, that is, attraction patterns in this language were also sensitive to the distributivity (the underlying notional plurality) of the preambles. This cross-linguistic disparity made people consider the possibility that not all languages behave in the same way. In her work in 1996, Vigliocco used the same type of completion task introduced by Bock & Miller (1991) in Spanish, but also some slightly different models: in some experiments, participants were presented with an adjective, then the preamble and then they had to assemble everything using a verb to link both parts. The results of this study replicated the ones already found in Italian in her work in 1995. When Vigliocco et al. found different results compared to those of Bock & Miller, they hypothesized that it was due to structural differences between the languages such as the word order (in Italian the subject of the sentence can be in a preor post-verbal position, while in English words normally follow a fixed order); that in Italian the subject can be omitted, and is, in fact, omitted in most of the declarative sentences; or that while in English the verb is inflected for person and number only in the 3 Agreement in psycholinguistics 33 third person singular in the present tense for regular verbs, in Italian the verb form is marked for number and person in every conjugation. These results led them to realize the need of “a model of grammatical encoding that allows for crosslinguistic variation” (Vigliocco et al. 1996:264). Later in this chapter this model proposed by Vigliocco et al. will be explained in detail. However, in the fourth and last experiment, Vigliocco et al. decided to approach this search for distributivity effects in English in a different way. This approach consisted in “looking for distributivity effects in English sentence types in which the produced utterance starts with the verb instead of the subject, such as questions” (Vigliocco et al. 1996:285). The word order that English follows in questions (subject in post-verbal position) should increase “the likelihood of independent retrieval of agreement features for the verb and the NP from the conceptual representation” (Vigliocco et al. 1996:287). The data resulting from this experiment confirmed the results of experiment 3 (Bock & Miller 1991): English speakers showed no sign of being sensitive to the notional number of the subject NP, regardless of the nature of the first element uttered (either the NP or the verb). So, summarizing the findings: Spanish, like Italian, showed distributivity effects, but English, as previously showed by Bock & Miller (1991) did not, even when producing utterances starting with the verb. These findings allowed the authors to reach two main conclusions with theoretical implications, which, as we have previously announced, are: languages react in different ways to distributivity due to structural differences among them; and that the findings support a model of “incremental grammatical encoding in which agreement is constructed via unification and no feature-copying” (Vigliocco et al. 1996:287). The structural differences among languages that make their agreement systems be affected in different ways are that they differ from one another regarding whether and where they permit semantic agreement; their morphology, that is, the richness of their verbal inflectional system, the word order (a fixed word order vs a more relaxed word order), or being a pro-drop language or not. These results support a unification based model because while a feature-copying approach to agreement would not “predict any semantic or pragmatic effects as the result of an independent retrieval of agreement features by verb procedure” (Vigliocco et al. 1996:290), a unification based approach would predict “a distributivity effect as a consequence of independent retrieval of agreement features from the conceptual representation at different points during incremental production” (Vigliocco et al. 1996:290). That is, the fact that some languages are sensitive to distributivity forces brought authors to propose a new model in which these interferences are possible, which is the unification model1. With these studies Vigliocco (1995; 1996) proved the interference of semantics, thus proving Bock et al (2001) wrong on this aspect. 1See 3.2.2: ”Maximal Input”. 34 3.2 Main models of agreement in psycholinguistics 3.1.3 Eberhard (1997) The third piece of work that changed the conception of agreement during the 90’s was the piece of work carried out by Eberhard (1997). This author undermined the cross-linguistic explanation proving that all the languages studied presented agreement operations sensitive to at least some semantic manipulations. For some time the English vs Romance contrast suggested by Vigliocco (1996) was thought to be a result of their belonging to different families of languages and therefore their already explained structural differences. However, Eberhard (1997) proved that, “when materials are carefully controlled for imageability, distributivity effects arise in English as well” (Acu˜ na Fari˜ na et al. 2014:112). In this research it was noted that it is the overall distributivity of the phrase what counts, not the collectivity of the local nouns (Bock & Miller (1991) had already reported that collectives do not attract in the local noun position), that is, words like army or team will not attract. Eberhard concluded that the main reason why Bock & Miller (1991) had not found any distributivity effect on their research was that the preambles were not correctly manipulated: they had low imageability. Before these results were reached it was believed that porosity (the interference of semantics in agreement) in the form of distributivity effects would be more common in the Romance languages due to their rich morphology, which provides opportunities “to consult meaning representations launched by the conspicuous cues” (Acu˜ na Fari˜ na et al. 2014:121). However, the predominant theory today is the opposite: a recent study by Lorimor et al. (2008) on attraction in richly-inflected languages, such as Russian, showed that this language was less affected by semantic interference than English. 3.2 Main models of agreement in psycholinguistics These main findings, results and corrections shaped the conception that psycholinguists had of agreement, how it works and what type of process it is. These different points of view were reflected especially in two main models, which are the ones that will be explained in this chapter. One of the psycholinguistic models that try to explain agreement is Marking and Morphing, a hybrid model formulated in 2001 by Bock, Eberhard, Cutting et al. It is called a hybrid model because it contemplates the possibility of both formal and semantic operations taking place at different moments in processing. This theory bears “a certain degree of encapsulation in its approach to the topic” (Riveiro Outeiral 2014:19). The other main psycholinguistic theory that will be discussed, Maximal Input, by Vigliocco et al. (2001; 2002), maintains that agreement can be influenced by semantic and/or pragmatic factors at all levels, thus, grammatical encoding can easily be affected by conceptual representations 3 Agreement in psycholinguistics 35 if that causes a processing advantage for the processor. In other words, they believe in a processor that takes into account all the available information from both conceptual structures and other sources, such as phonology, and that is able to incorporate them during all phases of processing. While Marking and Morphing is the theory ”heir to the inflectional, formal account of agreement” (Acu˜ na Fari˜ na 2012:14), Maximal Input presents the semantic, unification-based view of attraction. 3.2.1 Marking and Morphing The first model that will be discussed is Marking & Morphing. This model is heir to the first conception of agreement that resulted from the studies carried out by Bock et al. (1991), thus, as we have previously noted, it conceives agreement as a formally encapsulated process divided into phases that are strongly similar to the cyclic phases in linguistics. These phases that compose agreement are called marking and morphing; as we will see, each phase has a different function and is affected by different processes (the first stage is affected by semantics, while the second stage is on formal rails). Marking & Morphing “captures the notional influence that a parameter n has on the number value of a subject phrase, which comes directly from the first marking stage (the overall conceptualization of, for example, a collective as either a multiplicity of individuals or a whole)” (Acu˜ na Fari˜ na 2017:37). In other words, in the marking stage, where agreement starts, it is decided if the overall number of the subject phrase is singular or plural depending on “whether the notion expressed by the subject makes reference to a singleton or multiple things” (Acu˜ na Fari˜ na 2012:14). Thus, if the subject is table, the overall of the subject phrase will be marked as singular, while if the subject is tables, it will be marked as plural. This marking stage takes place during the mapping from message to a lexical-grammatical representation, and as we can see, we take into account the notional number, thus, semantics interferes in this stage. To this first marking, morphological calculations which are later weighted by hierarchical distance in the syntactic tree are added. This means that in a morphologically rich language a certain extra amount m is added to the notional component n. ‘M’ consists of two sub-components that are absent or diminished in languages with poor morphologies: a negative value when the subject head determiner is marked as singular, and a positive value when the interfering noun’s determiner is marked as plural. Since the model also factors in distance, in a morphologically rich language the negative value is presumably larger than the positive one, yielding more opportunities for the verb to be singular (Acu˜ na Fari˜ na 2017:37). The presence of the parameter M is what explains why languages with poor morphology are more prob- 36 3.2 Main models of agreement in psycholinguistics able to be affected by semantic interference and produce agreement mistakes, contrary to the general believe after the results of Bock et al. (1991). After presenting how the marking stage works, we can conclude that Marking is the mechanism that is in charge of transmitting the message number to the syntax –which carries number constraints from the message “in the form of features of an utterance’s structure” (Bock et al. 2005:8). As stated by Bock in (Bock et al. 2005:8): “The site of number marking is the seed or root of the subject noun phrase. Marking serves to make sure that, in the case of other information being absent, a subject noun phrase carries a number consistent with the notional number of the phrase’s referent. It can be singular or plural, or, presumably, null”. The idea that NPs might be notionally controlled is derived from the findings that show that noun phrases “may take plural agreement, despite having singular heads, when they support a highly accessible distributive construal” (Bock et al. 2001:89). This idea is reflected in examples that have already been mentioned like, for instance: the label of the bottles, where there is one label for each bottle, therefore, during the marking stage, the brain understands that there are several labels, thus, the phrase is plural. To this example of distributivity that supports the notional control of NPs, we can add other examples that we have already mentioned, but that are worth remembering: the many cases of semantic overrides that English is famous for2, such as number transparent nouns (a series of books were written), measure terms (three pounds is a ridiculous amount) or other types of agreement ad sensum (the team are thinking about the proposal). It is important to point out that “this message-based number marking is supposed to affect the subject, not the verb, at least in English” (Acu˜ na Fari˜ na 2012:14). This phenomenon, as we know, is a result of the impoverished English verbal system that provokes that verbs are considered unlikely to carry number indices (English verbs only have number feature mark in the third singular person: I sing vs she sings, or in the verb to be). This means that they acquire their number specifications through number Morphing (the stage after marking in which the analysis carried out in the first stage is passed on onto the verb). Marking is needed to deal with abstract phrase number (phrases with what,who, and which, which can be plural or singular), conjunctions (cold and wind), elided or lexically unrealized phrases (imperatives or expressions such as doesn’t matter) as well as differences in the number behaviour of pronouns and verbs (Bock et al. 2005:8). Due to structural characteristics of the English language such as a largely opaque meaning of the verb number, even to native speakers (as it is not morphologically marked in verbs apart from to be and most of the verbs in the third singular person in present), that the role of notional features in modern English agreement is weak, and that there is unlikely to be a direct assignment of number to the verb from the notions represented in the message, the linguistic requirements of verb agreement may be “more easily met by rooting the number features that support agreement in the notional referents of subjects alone and in English by providing access 2See 2.2.2: ”Unification”. 3 Agreement in psycholinguistics 37 to the notional underpinnings of grammatical number to the subject alone” (Bock et al. 2001:90). Except for “formally relevant features such as mass and count, which affect phrasal configurations, lexical-grammatical specifications of number do not come into play until the selected lexical entries are used in retrieving morphemes. This paves the way for number morphing” (Bock et al. 2001:91). After Marking, the following stage would be Number Morphing. As previously noted, in English it is the number marking of the subject function what controls the number features of the subject noun phrase. The assembling of the subject noun phrase takes place when the morphological representations of words and inflections are bound to specific positions within the phrase. “At this point, the specifications of number in the lexicon are morphologically instantiated and, if necessary, reconciled with the number features on the subject. This, along with the implementation of agreement operations, constitutes number morphing” (Bock et al. 2001:91). In English most of the morphemes do not carry intrinsic number specifications; this makes sure that in most of the cases a number marking that is notionally determined prevails. However, there are some morphemes which do have a number specification such as plural, singular or both simultaneously (for instance, the words tree and trees present a singular and plural morpheme, respectively, but words like fish do not have a morpheme to express their notional number). Therefore, “number specifications can mismatch the marked number of the phrase [which will trigger] an effort to resolve the marking of the phrase” (Bock et al. 2001:91). These number conflicts are often resolved by adjusting the number specification of the subject noun phrase in a stage that follows marking and morphing, but is not compulsory, it only occurs when it is needed, that is, when a mismatch has taken place. This adjustment (sometimes referred to as percolation in formal linguistics, or reconciliation) is “the compromise component of number morphing and it aligns morphological number and phrase number” (Bock et al. 2001:91). This stage is occasionally needed in cases where, as noted above, “the morphological number specifications of constituent nouns may mismatch the number of the whole noun phrase assigned –previouslyduring making” (Acu˜ na Fari˜ na 2012:15), as is the case of the already commented example of collective nouns: the team are playing. This type of conflicts is resolved by adjusting the number specification of the whole phrase, that is, by aligning morphological number and phrase number. The number morphing stage terminates in a number feature for the verb, which satisfies the requirements of number agreement, thus, constituting the control component of number morphing. The verb phrase acquires its number when it is transmitted from the subject noun phrase, and it is this “transmission what accounts for the findings regarding plural verb agreement with distributive subject noun phrases such as the label on the bottles” (Bock et al. 2001:92). The plural verbs in these cases, as argued by Bock et al. (2001), are a result of the notional plurality present in the message, and not of attraction. On the other hand, examples like The baby on the blankets are. . . , since they cannot be plural on the speaker’s mind, are considered as true attraction errors that take place via control. Attraction thus, “reflects the resolution of number conflicts between morphological number and phrase number” (Bock et al. 2001:91). 38 3.2 Main models of agreement in psycholinguistics The morphing stage proceeds on the basis of constraints imposed from the lexicon and syntax, and it helps to create representations that can support phonological encoding. So far we have seen how Morphing works, but now I will present the functions of this stage. The set of interrelated operations that conform the morphing stage have several functions, such as: (1) Bind morphological information to structural positions, (2) reconcile number-relevant features from the syntax (number marking) and the lexicon (number specification), and (3) transmit number features to structurally controlled morphemes (Bock et al. 2005:9). Summarizing: Morphing operates during structural integration [in order] to select and position the number morphology that [is represented in] pronouns and verbs. The marking and morphing account of verb morphing is straightforward: verbs inherit the number of the subject noun phrase. This constitutes agreement control of verb number by the subject number (Bock et al. 2005:9). The number of a subject noun phrase is a result (reconciliation) of “the initial marking of notional number and the number specifications of morphemes that compose the phrase [and that] capture the grammatical number properties of morphemes” (Bock et al. 2005:9). Therefore, it is a result of marking and morphing. The reconciliation of disparities between marked and specified number facilitates lexical specifications, in other words, when there is a disparity, it is easier for semantics to interfere. Summarizing this hybrid model: Marking and Morphing is a formal theory that conceives agreement as an encapsulated process divided into two different stages: Marking, affected by semantics, moment of the agreement procedure in which the overall number of the clause is decided; and Morphing, stage that occurs on formal rails, moment in which the analysis that took place during marking is passed on to the verb, with no semantic interference. 3.2.2 Maximal Input The following theory that will be discussed is a functionalist theory that presents the opposite agenda to Marking & Morphing. As noted above, the semantic, unification-based view of attraction is best represented today by the Maximal Input hypothesis of Vigliocco and collaborators (Vigliocco & Franck 2001; Vigliocco & Hartsuiker 2002). In essence, this theory maintains that: “even though agreement relations occur during a phase 3 Agreement in psycholinguistics 39 of morphosyntactic encoding, processing phases are not rigidly encapsulated, so that a previous conceptualisation may affect morphosyntactic operations if that is advantageous for the creation of an interpretation” (Acu˜ na Fari˜ na et al. 2014:112). This conception of phases as fluid rather than encapsulated allows the Maximal Input hypothesis to suggest that “the conceptual or notional information is directly used in implementing agreement”, (Bock et al. 2005:7) reinforcing the linguistic operations behind agreement and making the cases of agreement that are semantically supported less variable than the cases that are not semantically supported. According to Maximal Input “conceptual structures may exert a stronger control on the workings of the grammatical encoder. In this view, agreement processing is responsive to features of the conceptual representation beyond using them to assign syntactic features” (Riveiro Outeiral 2014:119). This theory assumes that every morph, that is, every agreement cue, is a symbol and that as such they come with a meaning side to be consulted. Maximal Input belongs to the group of theories that, as Constraint Satisfaction models and the Competition Model do, account for the influence of extrasyntactic factors in processing. These models share the view that a level of processing is not completely isolated, and can, therefore, experiment some interference from neighbouring levels, so, if circumstances make it advantageous for processing, grammatical encoding (the second level) may be affected by the previous conceptualisation stage and even by the following phonological stage (despite agreeing on this point, models differ in how much interference they allow). Maximal Input argues for more pervasive effects of meaning throughout the agreement process than they do in other theories of agreement. For instance, Vigliocco et al. predict that “notional correlates of local nouns should contribute to attraction” (Bock et al. 2005:19). This model allows to extend some of the elements of the Marking and Morphing theory to other kinds of agreement relations, other kinds of agreement features and other languages. For example, apart from the subject-verb and antecedent-pronoun number agreement, there also exists agreement in number between determiners and nouns. In English, there are words like a,each and this that occur with singular nouns (each kid,this person), while words like some,many and these, only occur with plurals (many children,these people). It might be assumed that these words, like personal pronouns, “carry number features that make them accessible for lexical selection from number meaning. This would [mean that] their agreement with the nouns they accompany [is] a matter of concord, as in the pronoun model, rather than control, as in the verb model” (Bock et al. 2005:19). In general, this theory presents a strong emphasis placed on the idea that “the grammatical encoder takes all of the information available from conceptual structures” (Acu˜ na Fari˜ na 2012:16). Nevertheless, Vigliocco & Frank (2001) make the careful observation that not all processing is subject to interference in an unconstrained way, since there are home domains for specific processes: Subject-verb agreement is computed during grammatical encoding, a stage in which the unfolding representa- 40 3.2 Main models of agreement in psycholinguistics tion is syntactic in nature, as suggested by slips of the tongue and other types of evidence. For example, Garret (1980) observed that errors occurring during this process [. . . ] do not show semantic/conceptual constraints. Instead, they show a strong grammatical class constraint (Acu˜ na Fari˜ na 2012:16). However, Vigliocco & Franck (2001), through the distinction of primary and secondary sources of information, incorporate the notion of interference, allowing some flexibility that is needed in to order clarify the complex facts of agreement. Maximal Input means that accuracy is achieved by taking advantage of information from other levels when available. Such information is ‘additional’ but nevertheless beneficial since it protects against information loss. On this view, when semantic information is congruent with grammatical information, agreement should be more accurate. (Acu˜ na Fari˜ na 2012:17). Nevertheless, this theory has been presented with a challenge that goes beyond the other challenges to constraint satisfaction: “the difference between verb and pronoun agreement in sensitivity to notional properties [. . . ] suggests that some kinds of processing must be more permeable than others” (Bock et al. 2005:22). This difference goes against what is proposed by Maximal Input. An explicit account of agreement would be needed in order to find out if more conceptual information is accessible to pronouns than to verbs, whether pronouns can capture differences in notional number better than verbs, or if it is something else what makes pronoun production more maximal than verb production. Marking and Morphing does offer explanations for these differences. Summarizing, this theory does not contemplate agreement as separated in encapsulated stages, but maintains that grammatical encoding can be affected by the previous, or even the subsequent, encoding stage, if this, for some reason, is advantageous. Thus, notional information can be acceded in later production processes, if this information is needed. Unlike in Marking and Morphing, interference is possible. Vigliocco et al. (2002) found support for the Maximal Input hypothesis in “the findings of conceptual influences on the construction of number and gender agreement” (Bock et al. 2005:7). Another interesting suggestion of this hypothesis is that languages with a rich morphology, like Russian, will have its agreement computations affected by semantic interference more easily than those languages with a poorer morphological system, like English. This opinion is opposite to the one presented by Lorimor et al. (2008), and mentioned before in this chapter, that it is in fact the languages with poor morphology the ones that will be more easily affected by semantic interference. 3 Agreement in psycholinguistics 41 3.3 Conclusion So, as a conclusion, and as we have previously said, the findings resulting from experimental work in production conducted in the 90’s, together with some corrections, shaped the way in which psycholinguists understand the process of agreement in the present days. Some of these different views of agreement are represented in the two theories that were presented in this chapter. As we have seen, they present two opposite points of view: while Marking & Morphing seems to be a more classic proposal, as it contains the “classic encapsulation that is characteristic of Fodorian views of language in general” (Acu˜ na Fari˜ na 2012:16), Maximal Input aligns itself with proposals that have been put forward in order to account for the influence of extrasyntactic factors and incorporates the notion of interference, thus allowing some flexibility needed to account for the notoriously complex facts of agreement. Despite decades and decades of research on the topic of agreement, authors have not been able to reach a conclusion on whether this is a semantic or a syntactic process, in fact, some authors (Acu˜ na Fari˜ na 2018) have understood this binary conception of agreement as problematic and have recently suggested the possibility of some sort of hybrid conception of agreement as a more fluid process than the more traditional idea. According to this proposal, agreement would be a dynamic system where depending on factors such as structural distance between controllers and targets, or the raw size of the morphological component, agreement would be more encapsulated or more permeable to conceptual pressures. With my study on agreement I do not intend to find the answer to this issue that has been tackled for decades by professionals, but to add to this field of investigation, especially on the feature of (semantic) concreteness, since there are not many studies on this topic. I will do so in the next chapter using the materials created by Riveiro Outerial (2014) through the method introduced by Bock & Miller (1991). 48 4.2 My study similar study which showed that concreteness did have an effect in agreement mistakes. Now, we will perform an experiment that is similar to the ones mentioned above. 4.2.1.1 Description Fourty preambles were taken from the seventy five that were created by Riveiro Outeiral (2014)1. The sentences used are presented in appendix A. 4.2.1.2 Objective The objective of the present test is to prove if the combination of an abstract singular head noun and a concrete local plural noun could produce more attraction errors in the systematic computation of agreement. If the processing of concrete words activates more areas of the brain and presents more advantages when compared to the processing of abstract words, this could mean that concrete words attract the verb when they appeared in their marked form (plural). This is the initial hypothesis for this test. 4.2.1.3 Method 4.2.1.3.1 Participants Twenty three people participated in the questionnaire. Most of them were students of high school level or university ranging from the age of 17 to 58. 69’6% of the participants were women and 30’4% were men. The first three people to do the questionnaire could not answer properly to three of the sentences due to a problem with the questionnaire (the questionnaire only allowed an answer with a number, which is not what we look for in this experiment), thus, they had to be eliminated and we had to add three more people in order to reach twenty participants with valid answers. None of the participants knew what the purpose of the experiment was. 4.2.1.3.2 Materials As previously mentioned in subsection (4.2.1.1), 40 preambles were taken from the 75 preambles created by Riveiro Outeiral (2014)2. These 40 preambles were distributed as follows: 1See apendix G in Riveiro Outeiral (2014) 2See apendix G in Riveiro Outeiral (2014) 4 My study 49 •Ten concrete singular – concrete plural (C-C) sentences formed the control condition. E.g. El campo de los elefantes. . . •Ten abstract singular – concrete plural (A-C) sentences formed the experimental condition. E.g. El significado de los besos. . . •Twenty fillers in which both nouns had the same number (either singular + singular or plural + plural). Since there is no possible semantic attraction in these filler sentences they will not be analysed as part of the experimental data. E.g. El due˜ no del gato. . . 4.2.1.3.3 Procedure Participants received a link to a questionnaire made on the online platform Google Forms [https://www. google.es/intl/es/forms/about/] through several social media such as Whatsapp, Instagram or email. They were presented with 40 incomplete sentences in written form that they had to complete with at least a formal verb and an adjective, also in written form. Due to our lack of resources we could not conduct the experiments as Bock & Miller (1991) and Riveiro Outeiral (2014) had done it, so we had to use this resource. 4.2.1.4 Results In the Spanish questionnaire there was only one agreement mistake in an abstract singular-concrete plural sentence. This means 1 out of 145 examples (145 after deleting 55 of the 200 because they were not correct, for instance, participants used a nominal phrase instead of an adjective), that is 0’69% of agreement mistakes in the A-C structure and, in this case, in the whole experiment. 4.2.1.5 Discussion These results reflect that concreteness interferes with subject-verb agreement in Spanish. The number of mistakes is quite low, but this is probably due to the procedure followed in this experiment. This provides data supporting the hypothesis that the structure [NP1 abstract + NP2 concrete] increases the possibility of making an attraction mistake on the verb. On the one hand, these results agree with the conclusion that was drawn from Bock & Miller’s (1991) research, that is, that marked nouns in a complex NP tend to produce more attraction on the verb, thus, making it agree with them. On the other hand, this contradicts another conclusion from Bock & Miller (1991): they had not found any effect of concreteness in agreement mistakes. In the next experiment to be discussed, the same difference in the number of agreement mistakes is expected to be found. Moreover, 50 4.2 My study as the morphological system of English is poorer than the Spanish system, a crosslinguistic difference is also expected, with the number of agreement mistakes in English expected to increase. 4.2.2 Second study: concreteness test (English) 4.2.2.1 Description The preambles used in the previous experiment3were translated into English and native speakers of English completed them. 4.2.2.2 Objective The objective of this test is the same as the previous one: to know if the combination of abstract and concrete nouns following the structure [Sg Abstract NP + Pl Concrete NP] would make the participants produce more agreement mistakes than with concrete nouns on both of the NPs. In this case, the language tested in the experiment was English. With this we intended to compare the results obtained in (4.2.1) with the experiments that we will obtain in this test, and also to draw a comparison between these and other results in the literature, such as the already mentioned results from Bock & Miller (1991) and Riveiro Outeiral (2014). 4.2.2.3 Method 4.2.2.3.1 Participants Twenty four native speakers of English participated in the questionnaire. 37’5% of the participants were men and 62’5% were women, ranging from the age of 19 to 67. The first four people to do the questionnaire could not answer properly to four of the sentences due to a problem with the questionnaire (the questionnaire only allowed an answer with a number, which is not what we looked for in this experiment), thus, they had to be eliminated and we had to add four more people in order to reach twenty participants with valid answers. None of the participants knew what the purpose of the experiment was. 4.2.2.3.2 Materials The same preambles used in the previous test were translated into English and distributed in the same way: 3See (4.2.1): ”First study: Concreteness test (Spanish)”. 4 My study 51 •Ten concrete singular – concrete plural (C-C) sentences formed the control condition. E.g. The field of the elephants. . . •Ten abstract singular – concrete plural (A-C) sentences formed the experimental condition. E.g. The meaning of the kisses. . . •Twenty fillers in which both nouns had the same number (either singular + singular or plural + plural). Since there is no possible semantic attraction in these filler sentences they will not be analysed as part of the experimental data. E.g. The owner of the cat. . . 4.2.2.3.3 Procedure Managing to find native English speakers who would fill in this questionnaire was more difficult than for the Spanish test. Again, we could not have them in a room to present them with oral preambles, so we had to use the resource of Google forms [https://www.google.es/intl/es/forms/about/] again. They received the link through Instagram, Facebook or via email. They were presented with fourty incomplete sentences that they had to complete with at least a verbal form of the verb to be in the past or present tense and an adjective. 4.2.2.4 Results Participants made 10 agreement mistakes out of 198 examples of concrete singular-concrete plural sentences (2 sentences had to be eliminated due to the fact that they did not follow the structure verb + adjective as requested in the instructions). In the abstract singular-concrete plural examples participants made 22 agreement mistakes out of 200 examples. That is 5% of agreement mistakes if a concrete noun is in the local noun position, and 11% if the local noun is abstract. 4.2.2.4.1 Parallel results As it has been previously explained, the structure of a complex NP that produces attraction mistakes is the one in which there is a mismatch following the pattern [NP1 sg NP2 pl]. However, while analysing the results of this experiment, we have found some unexpected cases of agreement mistakes. There were four examples in which there was not a number mismatch between the NPs, but which had eight answers (2% of the answers to the fillers) that presented agreement mistakes, as in *the cases for the pencils is green. All of these examples followed the structure [NP1 pl NP2 pl]. This type of results had already been found in other experiments, such as Riveiro Outeiral’s (2014) and it had been concluded that the repetition of an operation in a short period of time produces an effect on the parser, who tends to automatize this operation, that is, “the connection 52 4.2 My study (-s=plural) is automatized whenever it happens more than twice in a very short space of time, thus creating a visible priming effect” (Riveiro Outeiral 2014:205). This phenomenon appears to be increased in languages, like English, in which the mark for the plural in nouns is the same as the mark for the singular in verbs. 4.2.2.5 Discussion These results provide data supporting the hypothesis that the semantic force concreteness has some effect on agreement mistakes in English. This influence of concreteness on agreement procedures is much bigger in English than in Spanish, where it has certainly proved to be negligible. This is probably due to the lack of morphological marks, which forces the speakers to rely on semantic information, thus making the process a less automatic one (Berg 1998, Acu˜ na Fari˜ na 2009, 2012). Other languages, like Spanish, which present a richer inflection system allow speakers to pay less attention to the semantic information of each word, as phrasal packaging appears to run on the formation of alliterative sequences of the -0 -0 -0 or –A –A –A type. Moreover, these results show a disparity between the number of agreement mistakes when the head noun is abstract and when it is concrete. Once again, this supports the hypothesis that when a concrete noun is in the local noun position and an abstract noun takes the local noun position, the concrete noun attracts the verb. 5 CONCLUSION AND SUGGESTIONS FOR FURTHER RESEARCH In this piece of work information regarding the process of agreement has been presented: what it consists of, some special cases like Basque and other so-called “exotic” examples, and several terms that are central for the understanding of agreement processes. As well as the theories, both linguistic and psycholinguistic, that try to explain how it works from different perspectives; and then, specifically, the role of a semantic force –concretenessin this process. After this theoretical background, we conducted an experiment, which allowed us to add an experimental perspective to the theoretical study of agreement, and to reach a comprehensive understanding of this phenomenon. The results of the experiment have been analysed and we can draw several conclusions from them. First of all, this study confirms the hypothesis presented by Bock & Miller (1991) that in a complex noun phrase, the structure [NP1 sg + NP2 pl] produces agreement mistakes, while a number mismatch in a different order, or a non-mismatching structure, do not normally produce this type of agreement mistake. This surely tells us something about the architecture of the system. Moreover, we have found some evidence that indicates that semantics plays a role in agreement procedures. This is shown in the fact that in the case of complex noun phrases in which an abstract noun was in the head noun position, and a concrete noun in the local noun position, agreement mistakes were found twice as often than in the cases where both head and local noun were taken by a concrete noun, in the English experiment. This shows that participants found more trouble trying to process the abstract noun and therefore the concrete nouns attracted the verb to establish agreement with them. Thus, it proves that agreement is not a completely formal process, but that the semantics of the nouns matters. The most relevant results that allow us to reach these conclusions are the results drawn from the English questionnaire, as in the Spanish one there was only one agreement mistake. However, this crosslinguistic disparity was also expected and is relevant for our study. Furthermore, other non-expected results were found when analysing the data. As noted in the previous chapter, a small percentage of the answers given by the participants presented agreement mistakes even though both the local and head noun were in the plural, that is, the verb was in the singular (*The cases for the pencils is green). This has been called the “phonological priming” effect and it is a result of the repetition of the sound [s] across the complex NP, which makes the participant repeat it again in the verb, thus, using the 53 54 singular form is instead of the plural are:*The casesfor the pencilsis green. How do these results affect the way in which we see the hypothesis about the nature of agreement process? As we have seen, the results confirm the influence of semantics on agreement processes. Moreover, the results showed a crosslinguistic disparity resulting from the different morphological systems that English and Spanish present; thus, this proves that morphology also plays a role in the agreement processes. This crosslinguistic difference provides data supporting the theory presented by Lorimor et al. (2008) that holds that the richer the morphological system of a language is, the less affected the language will be by semantic interference. Since both semantics and morphology interfere in the agreement processing, the theories that conceive agreement as a completely encapsulated process should be discarded, as they do not contemplate the implications that these results provide. That is, theories such as Marking and Morphing (Block et al. 2001) are no longer adequate, according to these findings. There are some authors that, in the light of research that indicates the same as the findings of our study, propose a model in which the parser “takes advantage of the most valuable information in order to speed up the processing” (Riveiro Outeiral 2014:225). Acu˜ na Fari˜ na (2012) is one of the authors that support this type of model in his theory of “leaking and blocking”. In this theory it is argued that “the parser is able to select the most valuable information depending on the type of language (and the kind of morphological system the language has) and on the particularities of each and every structure” (Riveiro Outeiral 2014:225). This theory agrees with Lorimor’s (2008) proposal and it is the one that we support on this study. This paper is just a first step in the study of agreement and, more concretely, in the role that the semantic constraint of concreteness plays in its processing. Due to lack of time and resources this piece of research is not aimed to be exhaustive, as we could not reach to as many participants as it would be necessary in order to obtain more significant data. The same applies to the number of sentences that participants have to complete. Due to these and the following limitations, the results of this experiment should be interpreted carefully. Moreover, and this is the main problem that we faced while doing this research, making participants answer the questionnaire through Google Forms appears to be problematic. It seems that through this platform participants have too much time to think about the sentence, it is not as automatic as listening to the preamble, writing it down and then completing it. Thus, this results in fewer agreement mistakes than expected. For instance, in the Spanish questionnaire, there was only one agreement mistake out of approximately two hundred examples. Moreover, we also found that, even though it was specified in the instructions that the preambles should be completed with a verbal form and an adjective, in quite a lot of answers, especially in Spanish, we found that people answered with nominal phrases or just a verbal form with other adjunts, but not an adjective. Fortunately, this happened especially in the fillers, so we did not have to eliminate a lot of examples that were 5 Conclusion and suggestions for further research 55 relevant for the study, but the data could have been significantly reduced due to this situation. Thus, for further research in this realm a good course of action could be to ask participants to complete the sentences with the verb ser, as we did in the English questionnaire, and maybe put more than one example, so participants are completely sure of what we are asking. A APPENDIX: QUESTIONNAIRE USED IN THE CONCRETENESS STUDY IN SPANISH •C-C 1. El campo de los elefantes 2. El laboratorio de los investigadores 3. El patio de los ni˜ nos 4. La oficina de los investigadores 5. El uniforme de los trabajadores 6. La oficina de las chicas 7. La partitura de los compositores 8. La habitaci´ on de sus hijos 9. La ropa de los chicos 10. El sueldo de los trabajadores •A-C 1. El significado de los besos 2. El tama˜ no de las margaritas 3. La narraci´ on de sus fantas´ ıas 4. La causa de las victorias 5. La frecuencia de los orgasmos 6. La forma de las mu˜ necas 7. La importancia de sus caricias 8. El aspecto de las uvas 9. El peso de las estufas 10. El dise˜ no de las camisetas 57