Perception Verbs in Brazilian Portuguese: A Functional Approach

Disclaimer/Complaints regulations If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.


Introduction
The aim of this paper is to give a systematic description of the complementation patterns exhibited, both semantically and morphosyntactically, by perception verbs in Brazilian Portuguese within the framework of Functional Discourse Grammar (FDG, Hengeveld & Mackenzie 2008). To this end, we will first, in Section 2, give a brief outline of FDG. We then move to its treatment of perception verbs in Section 3. From this treatment a number of predictions follow, which are given in Section 4. The predictions concern the distribution of semantic complement types with perception verbs on the one hand, and the way in which the morphosyntactic complement types of perception verbs may be predicted from their semantics on the other. These predictions are tested in Sections 5 through 8. We round off the paper with our conclusions in Section 9.

Levels
The Interpersonal, Representational, and Morphosyntactic Levels of linguistic organization are constructed using different sets of primitives. Underlying the Interpersonal and Representational Levels of organization are pragmatic and semantic frames, which serve as hosts for lexemes and primary operators (operators that are defined in terms of their meaning). Underlying the Morphosyntactic Level are morphosyntactic templates, which receive, apart from lexical material from the preceding levels, grammatical words and morphosyntactic secondary operators (i.e. operators anticipating bound grammatical expressions). The Phonological Level is based on prosodic patterns, which host the lexical material handed over from the preceding levels, together with bound morphemes and possibly tertiary operators (i.e. operators anticipating the acoustic expression of the utterance).
Levels are related to each other through operations, represented with ovals in Figure 1. There is a crucial difference between Formulation on the one hand, and Encoding on the other. The process of Formulation deals with specifying the pragmatic and semantic configurations that can be encoded within the language. As far as Formulation is concerned, there may be differences between languages as regards the pragmatic and semantic functions that are necessary to describe their grammatical system. The process of Encoding deals with the morphosyntactic and phonological form pragmatic/semantic configurations take in a language. As far as Encoding is concerned, there may be differences between languages as regards their word order, phoneme inventory, morphological type, etc.
The levels that are most relevant for the current paper are the Interpersonal Level and the Representational Level. These will therefore be presented in somewhat more detail.
The Interpersonal Level is organized hierarchically as indicated in (1): (1) (M 1 : (A 1 : [(F 1 ) (P 1 ) S (P 2 ) A (C 1 : [...(T 1 ) (R 1 )...] (C 1 ))] (A 1 )) (M 1 )) The highest unit of analysis at the Interpersonal Level is the Move (M), which may contain one or more Discourse Acts (A). The central organizing unit within the Discourse Act is the basic Illocution (F), which takes the speech act Participants (P, the speaker S and the addressee A) and the Communicated Content (C) as its arguments. The Communicated Content itself is built up on the basis of a varying number of Ascriptive (T) and Referential (R) Subacts. The latter two units are operative at the same layer, which means that there is no hierarchical relation between them. The Interpersonal Level is thus an actional level, at which units are analysed in terms of their communicative function. The Representational Level is organized hierarchically as indicated in (2): (2) (p 1 : (ep 1 : (e 1 : (f c   1 : [(f l 1 ) (x 1 )] (f c 1 )) (e 1 )) (ep 1 )) (p 1 )) The linguistic units that are relevant at this level are categorized in terms of the semantic categories they designate. Propositional Contents (p) are mental constructs, only existing in the mind; Episodes (ep) are thematically coherent combinations of States-of-Affairs, where the States-of-Affairs show unity or continuity in time, space, and participants; States-of-Affairs (e) themselves are events or states, which have a temporal reality; Individuals (x) are concrete, tangible, entities that exist in space; and Properties (f) are special in that they only exist when they are applied to some other semantic category. Properties (f) occur both as units characterizing States-of-Affairs (the Configurational (c) Property (f c 1 ) in (2)), and as an independent unit (the Lexical (l) Property (f 1 )) within the Configurational Property. The units (f 1 ) and (x 1 ) in (2) are operative at the same layer, which means that there is no hierarchical relation between them.

Layering
Each level is organized hierarchically in terms of several layers. Higher layers contain lower layers. All layers at the Interpersonal and Representational Levels have the following general structure, where α ranges over all variables: (3) (π α 1 : [(complex) head] (α 1 ): σ (α 1 )) φ Lexical and grammatical means are used to build up each unit. Lexical means are the heads and optional modifiers (σ), where the head is shown as the first restrictor and the modifier as a non-first restrictor. Grammatical means are operators (π) and functions (φ). Operators specify non-relational properties expressed grammatically, functions specify relational properties expressed grammatically. The most important semantic domains of operators and modifiers for each of the layers of the Interpersonal and Representational Levels are given in Table 1. They are illustrated with examples of modifiers, as these will play an important role below.
In order to illustrate the above, example (5), adapted from Hengeveld & Wanders (2007: 217) is formalized interpersonally (5) and representationally (6) below: (4) Reportedly a man was slowly cutting himself with a knife yesterday. The different lexical modifiers (reportedly, slowly, with a knife, yesterday) are represented at their corresponding layers: reportedly modifies the C-layer at the Interpersonal Level in (5), deliberately, with a knife, and yesterday modify the f i -, f c i -, and ep i -layers of the Representational Level in (6). An aspectual and a temporal operator at the f c i -and ep i -layer in (6) trigger the past progressive form of the verb.

Complementation
In the preceding paragraphs it was shown that layers may be qualified in similar ways by either operators or modifiers. In the same way, they may be used as arguments of complement-taking predicates, and these then again qualify the layers they take as their arguments in similar ways as operators and modifiers.
Thus, complement-taking verbs with meanings parallel to the ones listed in Table 1 for the Interpersonal and Representational Levels, take arguments with systematically decreasing internal complexity the lower the specific layer they embed. As a consequence, the layers listed in (7) and (8), taken from Hengeveld & Mackenzie (2008, chapter 4), may occur as arguments of complement-taking predicates: Interpersonal layers underlying subordinate clauses a (Π M 1 : ): σ (f c 1 )) (e 1 ): σ (e 1 )) (ep 1 ): σ (ep 1 )) (p 1 )) b (π ep 1 : (π e 1 : (π f c 1 : : σ (f c 1 )) (e 1 ): σ (e 1 )) (ep 1 ): σ (ep 1 )) c (π e 1 : (π f c 1 : Lower layers are contained within higher layers. Therefore, subordinate clauses may be classified in terms of the highest layer they contain. In addition, as a layer always brings along its particular set of operators and modifiers, it may be predicted that all the operators and modifiers qualifying the highest layer underlying a certain type of subordinate clause, and all lower operators and modifiers, may be expressed in such a subordinate clause. On the other hand, modifiers and operators qualifying layers higher than the highest layer underlying a certain type of subordinate clause, are excluded from expression in such as subordinate clause. The following examples (see also Hengeveld & Mackenzie 2008: 361-367) illustrate this for the Interpersonal Level: While it is difficult to make generalizations about such a diverse public, it is easy to conclude [that in sum, these actions have led to a net loss of vegetative cover relative to pre-settlement conditions, as well as a substantial change in the type of vegetation present. At the same time, public consciousness regarding the importance of urban vegetation has certainly risen in the last ten years, although how much of that awareness has translated into changed behavior vis a vis urban plants in The complement-taking predicate conclude in (9) takes a summarizing Move in an argumentative series as its argument. This Move is represented as the Undergoer (U) argument M 1 of the verb conclude and is itself built up as a series of Discourse Acts. We can now explain the presence of the modifier in sum in the subordinate clause, as this is an M-modifier. The complement-taking predicate add in (10) has a single Discourse Act A 1 as its argument, and this explains why it is possible that the A-modifier frankly speaking but not an M-modifier may be expressed within it. Finally, the complement-taking predicate state in (11) takes the Communicated Content C 1 as its argument, and therefore may contain the C-modifier reportedly, while the presence of an A-modifier or an M-modifier is not grammatical. : [(f 2 ) ...]) (f c 1 )) (e 1 )) (ep 1 )) (p 1 ): σ (p 1 )) U (13) It ends with him (*possibly) breaking up with her and her running crying out into the hall when he returns from the war. The complement-taking predicate believe in (12) is the expression of a propositional attitude, and its argument therefore has to be a propositional content (p 1 ). For this reason it may contain the modifier possibly, which expresses a propositional attitude. The complement-taking predicate end_with in (13) takes an argument (ep 1 ) that denotes the final episode of a story. It may therefore contain an absolute temporal modifier, in this case when he returns from the war, but a p-modifier is not grammatical. The complement-taking predicate see in (14) takes the witnessed State-of-Affairs (e 1 ) as its argument. This explains the presence of the relative temporal modifier before dinner, and the fact that p-modifiers and ep-modifiers are not grammatical, at least not as modifiers of the subordinate clause. Finally, the complement-taking predicate continue in (15) denotes the persisting presence of a property, and it therefore takes a Configurational Property (f c 1 ) as its complement. The presence of this layer licenses the modification by the aspectual adverb uninterruptedly, while in the absence of higher layers p-modifiers, ep-modifiers, and e-modifiers are not grammatical, again when these are taken as modifiers of the subordinate rather than the main clause.
We conclude, then, that it is possible to classify complement clauses on the basis of the highest layer they contain. Since lower layers are included in higher layers, the presence of the highest layer predicts the presence of all lower layers as well as the operators and modifiers corresponding to them.

Introduction
Perception verbs, like see and hear, specify a relation between an individual (the perceiving entity) and different kinds of the representational/interpersonal categories introduced above, according to the nature of what is perceived. In Dik and Hengeveld (1991), a description of the different kinds of perception verb complements is given within the Functional Grammar framework, accounting for the many subtle semantic differences between them. Drawing on earlier work by e.g. Kirsner & Thompson (1976), Holierhoek (1980), Barwise & Perry (1983), Noonan (1985), and van der Auwera (1985), the authors argue that perception verb complements can be understood in terms of the hierarchical clause structure used in Functional Grammar to represent utterances. In the same way, in this section we carry out the characterization of perception verbs and their complements, but now according to the FDG model, taking the previous description as our starting point. We will show that perception verbs can take five different types of complement: Properties (f), Individuals (x), States-of-Affairs (e), Episodes (ep), and Communicated Contents (C).
From this section onwards we will use Brazilian Portuguese examples. All these examples were obtained through internet searches using the Google search engine. They were subsequently checked for their grammaticality by the three authors of this paper that are native speakers of Brazilian Portuguese.

Perception of Property
A perception verb in this case describes the perception of a property by an individual. Since properties do not exist by themselves, the object of perception is a characteristic of another entity, as illustrated in (16) The verb sentir 'sense' in (16) specifies a relation between the perceiving Individual (x), the understood subject 'they', and the perceived Property (f) o cheiro de comida estragada 'the smell of spoiled food'.

Perception of Individual
In this use of perception verbs what is described is the perception of one individual by another, as illustrated in (17) In this case, the verb ver 'see' specifies a relation between two semantic categories of the same type: a perceiving Individual (x) eu 'I' and a perceived Individual (x) o passarinho 'the little bird', both individuals, and, consequently, concrete and tangible entities.

Perception of State-of-Affairs
This reading concerns the direct perception of a state-of-affairs by an individual, as the following example shows: In (18), the verb ver 'see' specifies a relation between the Individual (x) category eu 'I' and a directly perceived State-of-Affairs (e) um carro batendo numa bike 'a car crashing into a bicycle'.

Perception of Episode
The fourth possible reading concerns the deduction of a piece of knowledge by means of perception through one of the senses, as illustrated in the following sentence: As is clear from the tenses used, in this example the first person subject did not witness a car crashing into a bicycle directly, as in (18). Rather, he/she comes to the conclusion that the crash has taken place on the basis of visual evidence. The difference with (18) is that in (18) the complement clause represents the stateof-affairs witnessed directly and is thus of the e-type, while in (19) it represents the conclusion the speaker arrived at. Dik & Hengeveld (1991) call this type 'perception of propositional content'5. We here choose, however, to classify it as the perception of Episodes, following Hengeveld & Hattnher (2015). These authors situate the evidential category of deduction at the layer of the Episode, on the basis of the fact that 'deduction necessarily involves at least two related states-of-affairs: the perceived one and the deduced one. The speaker deduces the occurrence of one state-of-affairs, the deduced one, on the basis of another state-ofaffairs, the perceived one' (Hengeveld & Hattnher 2015: 486). As it is within the Episode that the relation between States-of-Affairs is specified, deduction must then be situated at that layer.
The connection between the two States-of-Affairs within the Episode is also shown in the fact that there has to be a (relative) temporal connection between the perceived and the deduced events, as illustrated in (20) (Hengeveld & Hattnher 2015: 490-491): a. I smell that he has been cooking. b.
*I smell that he had been cooking.
The temporal specification in the complement clause in (20a) expresses relative tense, which connects the perceiving event with the deduced event. In (20b) the complement clause contains an expression of absolute tense, and thereby disconnects the perceiving event from the deduced event, which leads to ungrammaticality. Given the requirement of a temporal connection, the two events must be within a single Episode.
There are a number of grammatical differences between constructions that express the perception of a State-of-Affairs and those that express the perception of an Episode that allow us to distinguish them, as shown in Dik and Hengeveld (1991). These are: (i) the simultaneity of the e-complement with the main clause; (ii) the impossibility to negate the e-complement independently; and (iii) non-factivity, i.e., the absence of a presupposition on the part of the speaker that the e-category took place.
The first property is shown in (21) While the use of the past tense is fine in (19), it leads to ungrammaticality in (21). This is because direct perception requires simultaneity of the perceiving and the perceived State-of-Affairs.
The examples in (22)  When the verb ver 'see' takes an Episode as its complement, it describes acquisition of knowledge. Predicates of acquisition of knowledge are semi-factive, that is, the speaker presupposes that the complement describes a fact. For this reason, the continuation in (23b) is ungrammatical.

Perception of Communicated Content
This reading is only possible with predicates of hearing and seeing (in the sense of 'reading') when used by the speaker to relay words or thoughts of someone else, as illustrated in (24) In (24), the verb ver 'see' specifies a relation between the first person singular perceiving x-type subject 'I' and the perceived Communicated Content que um jovem de 21 anos matou o irmão de 22 'that a 21 years old boy killed his brother of 22', which represents a piece of information claimed by a third party.
The grammatical expression of the perception of a Communicated Content is different from that of an Episode. As shown in (24), in the former case the source of the information, here o jornal 'the newspaper', may be specified. In the latter case, this is not possible, as shown in (25)

The representation of perception verbs in FDG
Constructions with perception verbs, like the ones presented in 3.2 to 3.6, are used to express the subject's perception of an aspect of the extralinguistic world. In this way, this kind of construction is dealt with at the Representational Level in the FDG model.6 In this subsection we will present the underlying representations for the constructions with perception verbs presented so far. As shown before, perception verbs can have different representational or interpersonal categories as their complement, and these are represented by different variables. We may therefore formalize the differences between them exploiting the variables introduced earlier. 'PV' is shorthand for 'perception verb'.
(i) perception of Property: (ii) perception of Individual: (iii) perception of State-of-Affairs: In the representations from (26) to (29), the variables representing the perception verb complement pertain to the Representational Level. In (30), it belongs to the Interpersonal Level. This is due to the fact that, in this reading of perception verbs, the complement of the verb is the Communicated Content produced by a different speaker in an interpersonal act.

Predictions
After introducing the theoretical background and the classification of the complement types of perception verbs that follows from it, we now may formulate two predictions concerning the distribution of semantic complement types and their morphosyntactic expression. Not all perception verbs may occur with all five semantic complement types introduced above. For instance, the verb ver 'see' was used above to illustrate all five complement types, as it is compatible with all of them. Other verbs, however, such as provar 'taste' have a much more limited range of possibilities. This particular verb only occurs with f-complements and x-complements, shown in (31) The question is now whether there is any systematicity in the distribution of semantic complement types across perception verbs. We expect that there is. Our prediction is that it is likely for perception verbs to take complements based on lower layers, while it becomes more unlikely for them to take complements based on higher layers. The reason is that basic perception is a physical process, and that the higher one gets in terms of layering, the less concrete and the more abstract the layers become. We thus predict that individual perception verbs will take semantic complement types according to the following implicational hierarchy: That is, if a certain perception verb allows a complement of, say, the ep-type, it will also allow all the complement types to the left of ep in the hierarchy. And if it does not allow, for instance, a complement of the x-type, it won't allow all the complement types to the right of x either. There may be a diachronic Brought to you by | Universiteit van Amsterdam -UVA Universiteitsbibliotheek SZ Authenticated Download Date | 3/3/20 10:21 AM dimension to this as well, as it might be that perception verbs start out with lower layer complements and expand the range of complements over time passing along the hierarchy.
Turning now to the morphosyntactic expression of perception verb complements, the question is whether we can also predict how the different morphosyntactic types of complement are distributed across the different semantic types. As has become clear in the various examples shown above, complements may take the form of noun phrases, non-finite clauses (infinitival and progressive), and finite clauses. As shown in earlier work (Hengeveld 1998), the higher the layer a subordinate clause contains, the more likely it is to be expressed by a finite construction. The reason for this is that, as the number of layers increases, the number of grammatical categories to be expressed also increases. We may thus expect the following mapping between the semantic types of complement represented in (32) and their morphosyntactic expression: As f-complements and x-complements may only be expressed by noun phrases, they are not relevant categories to test this prediction. For the remaining types of semantic complement (33) predicts two things. First, finite complements are more likely to be found to the right of the hierarchy and non-finite complements are more likely to be found to the left of the hierarchy. And secondly, when a category to the right in the hierarchy is expressed by non-finite forms, then the categories to the left of it are also expressed by these forms; and when a category to the left in the hierarchy is expressed by finite forms, then the categories to the right of it are also expressed by these forms. In the following we go into the semantics of perception verbs and their complements in Section 5 before testing the first prediction in Section 6. We then describe the morphosyntax of perception verb complements in Section 7, and test the second prediction in Section 8.
Especially remarkable is that in three cases the same verb may be used to express visual and auditory perception. A typological study by Viberg (1984) shows that it is uncommon for languages to not express visual perception by a separate lexical item, a situation which occurs in only three of his 53 languages. In none of Viberg's three cases does the polysemy8 exhibited concern just visual and auditory perception. Aikhenvald & Storch (2013: 16) already noted a number of cases like these, and in Brazilian Portuguese this type of polysemy is found as well, though it is restricted to the perception of properties. Examples (34)-(36) show the use of ver 'see', observar 'observe', and notar 'note' to express visual perception: Given the extensive amount of polysemy observed, where necessary we will indicate with a superscript which reading of a perception verb is intended. Thus ver A will mean that the verb ver 'see' is used in its auditory reading.9

The distribution of semantic complement types
In Section 4 we predicted that perception verbs take different sets of semantic complement types according to the following hierarchy: For the question of polysemy in perception verbs, see also the discussion in Gisborne (2010). 9 The abbreviations used are A for auditory, G for gustatory, O for olfactory, T for tactile, and V for visual. This hierarchy predicts that semantic complement types more to the left of the hierarchy are implied by the presence of semantic complement types more to the right of the hierarchy. Table 3 shows that this prediction is fully borne out by the data. The data on which this table is based are all given in Appendix 1. Note that a '+' in Table 3 indicates that a particular complement type is attested, while a blank indicates that it was not attested.
At the top of Table 3 the perception verbs with the widest range of semantic complement types are given, at the bottom those with the narrowest range are given. The verbs at the top combine with all possible complement types, the ones at the bottom only with the property denoting complement type, the lowest one on the hierarchy. All intermediate cases show systematic decreasing combinatorial possibilities following the various steps in hierarchy (40).
The verbs highest on the hierarchy are verbs of visual and auditory perception taking a C-complement. This is not surprising, as linguistic units can only be perceived through reading and listening, i.e. through visual and auditory perception. At the other end of the hierarchy we find verbs with a primary visual reading being used in an auditory sense. We do not see an evident explanation for this fact. In between we find other sets of combinations of perception verbs with semantic complement types, but importantly these always obey the hierarchy in (40). Our first prediction is thus fully borne out.

The morphosyntax of perception verbs in Brazilian Portuguese
The perception verbs analyzed in this paper allow various types of morphosyntactic complement. The first division is between noun phrase complements and clausal complements. Within the group of clausal complements we find finite and non-finite clauses, and the latter group consists of progressive and infinitival clauses. The other way round, complements designating a Property or an Individual cannot be expressed by clauses. They can of course be expressed by headless relatives, as in (51)

The distribution of morphosyntactic complement types
In Section 4 we predicted the following distribution of morphosyntactic complement types with perception verbs in Brazilian Portuguese: This prediction follows from the idea that the higher the layer on which the semantic complement type is based, the more likely it is that this complement type will be expressed by a finite complement clause. The first two categories in the hierarchy are irrelevant for this prediction, as they can only be expressed by noun phrases. As shown in the previous section, not all perception verbs allow all semantic complement types. In order to test the prediction in (52) we therefore have to limit ourselves to the perception verbs that allow a wide range of semantic complement types. The ones we have selected are those that allow at least an Episode as their semantic complement. Table 4 shows which perception verbs comply with this criterion. The data on which this table and later tables in this section are based are all given in Appendix 2.  Table 5 now shows the ways in which the complement types expressing States-of-Affairs, Episodes, and Communicated Contents are realized morphosyntactically in terms of finiteness. In this table a + indicates that a complement is finite, a -that it is non-finite, while 'irr' indicates that a slot is irrelevant. What is clear from Table 5 is that there is a clear split between complements designating States-of-Affairs on the one hand, and those designating Episodes and Communicated Contents on the other. The former are always expressed through non-finite forms, the latter through finite and non-finite forms. A further generalization arises when we further distinguish between the two non-finite forms, progressive and infinitival forms, and consider their distribution across semantic complement types. This is shown in Table 6.  Table 6 shows that progressive forms are only found in the expression of complements designating Statesof-Affairs. A construction with an ep-or C-complement in the progressive form is ungrammatical, as shown in (53), or results in a State-of-Affairs reading, as in (54) Together with the data in Table 5, this leads to the overall picture presented in Table 7. In all, and as predicted, we thus see a clear relationship between the semantic complement types on the one hand, and their morphosyntactic expression on the other.11

Conclusions
In this paper we have shown that the complements of perception verbs in Brazilian Portuguese can be classified semantically using the semantic and pragmatic categories proposed in Functional Discourse Grammar. Complements of perception verbs can be argued to express Properties (f), Individuals (x), Statesof-Affairs (e), Episodes (ep), and Communicated Contents (C). This subdivision into complement types is relevant in two different respects. First of all, the set of semantic complement types that a perception verb can take is not random but follows a hierarchy, in which the categories mentioned above are ranked from lower to higher scope. If a perception verb can take a semantic complement type of a certain scope, it can also take all other semantic complement types with lower scope. And secondly, the morphosyntactic expression of complements of perception verbs in Brazilian Portuguese is closely linked to their semantic types: the higher a complement in the semantic hierarchy, the more likely it is to be expressed by finite forms. We furthermore found that progressive forms are limited to complements denoting States-of-Affairs. In all, this study thus has shown that the semantic categories of complements distinguished in FDG provide a useful categorization that helps to systematically describe the semantic and morphosyntactic behaviour of perception verbs in their many readings as well as their complements in their many formal manifestations.