1.1 Background

Do people around the world think about sensory perception in the same way? On the one hand, we might think they do, given that people everywhere use the same sense organs to perceive the world around them. On the other hand, we may expect differences, given that the role of the different sensory modalities in daily life depends on cultural practices, lifestyle and the environment in which people live. One way of approaching this question is by investigating how people talk about perception in different languages cf.(Majid and Levinson 2011). What are the similarities and differences among the languages of the world in their vocabularies and grammatical constructions for describing perceptual experience?

This paper contributes to answering this question by focusing on a subset of perception vocabulary: the basic verbs that describe perceptual experience, such as English see and hear. It also zooms in on one language: Avatime, a Ghanaian Kwa (Niger-Congo) language.^[1] Avatime has two verbs to refer to perceptual experience: one for vision (mɔ̀ ‘see’) and one for the other senses (nu ‘hear/feel/taste/smell’).^[2] The aim is to investigate what this particular verbal encoding pattern could mean for how sensory perception is conceptualized. More specifically, two aspects of the language of perception are investigated: (i) how are the sensory modalities encoded in verbs and grammatical constructions, and (ii) what other meanings are encoded by the basic perception verbs. I will briefly introduce each of these two aspects in turn.

Typological linguistic work investigating the encoding of the sensory modalities in verbs was pioneered by Viberg (1983). He observes that languages differ in their number of basic perception verbs. Whereas some languages, like English, have a verb for each of the five senses (vision, hearing, touch, taste, smell), other languages combine multiple sensory modalities in a single verb. Viberg assumes that such conflations of multiple sensory modalities come into existence through meaning extensions, where a verb that originally refers to a single sensory modality acquires additional meanings. He observes that these meaning extensions follow a hierarchy with vision at the top, then hearing, then touch, then taste and smell. This hierarchy has also been attested for perception verbs in Australian languages (Evans and Wilkins 2000), but counterexamples have also been found see e. g.,(Aikhenvald and Storch 2013).

Viberg (1983) assumes that perception verbs that encode multiple sensory modalities are polysemous and that the traditional five senses are the basic concepts of perception to be lexicalized in the languages of the world. However, anthropologists of the senses have argued that the five-senses model is a Western cultural construction see e. g.,(Classen 1993). Other cultures may have different ways of thinking about sensory experience. They could for instance view the mind as a sixth sense (Classen 1993: 2) or conceive of a sense of balance (Geurts 2002). From this point of view, the perception verbs of a language can provide insight into a culture’s conceptualization of the senses. For instance, Howes (2006) observes that speakers of several Melanesian as well as African languages talk about ‘hearing a smell’ and connects this to what he calls audio-olfactory synesthesia, i. e., the cultural conflation of the two sensory modalities. Ritchie (1991) claims that speakers of Hausa distinguish only two sensory modalities, based on the observation that Hausa has only two perception verbs, one for vision and one for the non-visual sensory modalities. So where Viberg (1983) assumes a disconnect between the encoding of perceptual experience in verbs and people’s conceptualization of the senses, these anthropological studies assume a direct link.

For the case of Avatime, which, like Hausa, has one verb for vision and one for the other sensory modalities, this means that different predictions could be made. One, assuming a direct link between perception verbs and conceptualization, is that the Avatime non-vision verb corresponds to a holistic ‘non-visual’ sensory modality. The other, assuming a complete separation of verbal encoding and conceptualization, is that the Avatime verb is polysemous with four separate senses (‘hear’, ‘feel’, ‘taste’ and ‘smell’). Alternatives in between these two extremes are also possible. The current paper investigates which of these predictions holds true.

The second aspect of the language of perception mentioned above is the additional meanings of perception verbs outside the domain of perception. This topic was put on the agenda by Sweetser (1990), who observed links between particular sensory modalities and particular meanings in the domain of cognition in a number of Indo-European languages. Sweetser (1990) explains these links by referring to specific properties of our sensory modalities that can be mapped onto cognitive processes. She notes that vision is connected to the intellect, with vision verbs coming to mean ‘know’ and ‘understand’. This can be explained by several aspects of vision: (i) vision is the main sensory modality by which we acquire objective knowledge about the world (ii) we can focus our vision on different elements of the environment, just like we can focus on different problems with our minds, (iii) vision is a distant sense, just like objective knowledge is considered distant and impersonal, and (iv) we assume that what different people see is the same and therefore knowledge acquired by vision is objective. Hearing, being the main way in which most people perceive language, is linked to the reception of speech, with meanings such as ‘obey’ and ‘understand’. Feeling and tasting are linked to more subjective notions such as emotion, personal preference and opinion, because they are personal and proximal senses.

Further research on a more diverse variety of languages has shown that the links between perception and cognition are more culturally variable than Sweetser (1990) envisaged. In a number of languages around the world, hearing is the sensory modality most closely connected to knowledge and the intellect (Aikhenvald and Storch 2013; Evans and Wilkins 2000; Ibarretxe-Antuñano 2008; Vanhove 2008). Evans and Wilkins (2000) speculate that, for Australian languages, this could be related to the culture of the speakers, in which hearing is considered a more focused activity and more important for the acquisition of knowledge. These further findings are compatible with the idea of a potentially universal general link between perception and cognition, with culture-specific mappings from particular sensory modalities to particular cognitive notions (Ibarretxe-Antuñano 2008; San Roque et al. 2018).

Coming back to Avatime, what kinds of non-perception meanings can we expect for the two perception verbs mɔ̀ ‘see’ and nu ‘hear/feel/taste/smell’ and what can these tell us about the speakers’ conceptualization of perception? Given that both vision and hearing have been attested to be linked to objective knowledge, both Avatime perception verbs could potentially have additional meanings in this domain. We can also expect the verb nu to be used in the domain of emotions and personal preference, given that it encompasses the personal and proximal notions of ‘feel’ and ‘taste’. Previous research into other general non-vision verbs in Romance languages (Enghels and Jansegers 2013; Guerrero and Cruz Domí nguez 2017) and Lussese (Bantu) (Thanassoula 2013, 2016) has shown these verbs to have a wide variety of non-perception meanings, referring to beliefs, emotions, opinions, knowledge and understanding. This could indicate that the domain of cognition is linked more strongly to the non-visual senses than to vision in these languages. Whether this is also the case for Avatime will be investigated in this paper.

In order to shed light on the issues mentioned above, a detailed analysis of the grammatical properties, semantics and discourse usage of the Avatime perception verbs is necessary. More specifically, the following questions will be addressed.

1. How is the non-vision verb nu used within the grammar of the language? Do the grammatical constructions in which it occurs disambiguate the different interpretations of this verb?

2. Is the non-vision verb nu polysemous or semantically general and do its interpretations map onto the traditional five sensory modalities?

3. What interpretations outside the domain of perception do the vision verb mɔ̀ and the non-vision verb nu have? What can these tell us about how perception is conceptualized?

4. How do people use the two verbs to talk about perception in discourse? To what extent are they used to refer to perception in general and to perception with specific sensory modalities?

These questions will be addressed in the next four sections, after a discussion of the method and data.

1.2 Method and data

The research questions mentioned above will be addressed using two different methods: translation of English sentences to Avatime and the analysis of more spontaneous discourse.

For the translation task, I largely followed the Grammar of Perception Questionnaire (Norcliffe et al. 2010). This questionnaire was developed to investigate how meaning is packaged into grammatical constructions in perception event descriptions and what differences there are in the grammatical means used for the different sensory modalities. It consists of 60 main scenarios and 63 optional additional scenarios that tap into various semantic aspects of perception events that may be relevant for grammatical encoding. Each scenario consists of a context part and a target part that is to be translated by the language consultant. An example can be seen in (1), where the part in bold is the target part and the context makes sure that the English word taste is interpreted correctly (in this case as referring to active tasting rather than the passive experience of taste).

All main scenarios and some of the optional scenarios were translated by six speakers of Avatime.^[3] A number of additional scenarios relevant to Avatime were developed and translated by at least two consultants each.

The translation of questionnaire scenarios was complemented by the study of perception verbs in spontaneous discourse. The corpus of discourse used for this study consists of transcribed and translated recordings collected over the course of several field trips to the Avatime area between 2008 and 2016 by Rebecca Defina and myself. The genres included in the corpus are: elicited narratives (based on pictures or videos), traditional narratives, life stories, descriptions of procedures, interviews, focus group discussions, public meetings and everyday conversations.^[4] There are a total of 76 recordings that contain the verbs mɔ̀ or nu. The total length of these recordings together is over 15 hours. The corpus contains a total of 223 occurrences of nu ‘hear, perceive’ and 839 occurrences of mɔ̀ ‘see’. All occurrences of nu and approximately 70% of the occurrences of mɔ̀ were coded for their meaning in context, the type of object they occur with and the grammatical construction they occur in.^[5]

Whereas the translation data provides an overview of the grammatical constructions that can be used for situations which may be rare in spontaneous discourse, the corpus data adds information on both frequency and additional uses of verbs that may not have been covered by the preconceived situations presented in the questionnaire. Together, the two methods allow for a comprehensive and detailed picture of the grammar and use of perception verbs in Avatime.

2.1 Nu with a perception object

Unlike English, which allows any noun phrase as the object of its perception verbs, the Avatime verb nu imposes a semantic restriction on its objects. It can only take objects which refer to an abstract perceptual signal. These are nouns such as òzi ‘noise’, ɔ̀dẓi ‘song’, ɔ̀f̣ù ‘scent’ and kuhuhu ‘heat’. Examples of nu with a noun referring to such a perceptual signal are shown in (2) and (3).^[6] Nouns referring to other types of referents (e. g., baṇùvɔ̀ ‘children’, òhui ‘car’, séfofo ‘flower’ and òni ‘soup’) cannot directly occur as the objects of nu, as illustrated in (4) and (5).

If one wants to talk about non-visually perceiving concrete referents such as a car or soup, there are two possible constructions. The first is to combine this concrete referent with the relevant perceptual signal in a possessive construction. This complex NP can then be used as the object of the verb nu. This strategy is conventionally used for the sensory modalities of smell and taste, using the nouns ɔ̀f̣ù ‘scent’ and òle ‘taste’ respectively. An example can be seen in (6). Here, the NP kùkwiyɛ ‘the pepper’ refers to a concrete perceived referent and is combined with the NP òlelò ‘the taste’, which refers to the perceptual signal. Possessive constructions in Avatime consist of juxtaposition with possessor-possessum word order, so the complex NP kùkwiyɛ òlelò means ‘the taste of the pepper’.

The second type of construction for talking about a concrete perceived referent uses the postposition ònu. This construction is used mostly for the sensory modalities of hearing and touch, as exemplified in (7) and (8).

The nu + ònu construction can also be used for smell, but this is rare. One example can be seen in (9). Whether or not the construction can also be used for the sensory modality of taste needs further investigation, but there are no cases of this in the questionnaire or corpus data.

The postposition ònu is also used for spatial expressions, to indicate a location at the opening or entrance of a reference object. An example of this usage can be seen in (10).

It is unclear how the spatial meaning of ònu is related to its meaning in perception constructions. A possible route is via sentences such as (11). Here, the postposition ònu indicates a metaphorical location – the source of information. This is similar to the use of the English spatial preposition from. However, unlike English from, the postposition ònu is not normally used in spatial expressions to indicate a source of motion, so there is no direct link between the spatial and perception usages of ònu.^[7]

Based on the data presented so far, we can conclude that some of the meanings of the verb nu can be disambiguated by the grammatical construction in which the verb occurs. To talk about taste and smell, the noun phrase describing the perceived referent is usually combined with the nouns òle ‘taste’ and ɔ̀f̣ù ‘scent’, keeping these two sensory modalities apart. Hearing and feeling by touch, on the other hand, can only be referred to with a more general construction. This raises the question if and how people distinguish hearing and feeling if they have to. To investigate this, I asked four consultants to translate sentences such as ‘he did not hear it but he felt it’. An example of a response can be seen in (12). Here, the speaker uses the nu + ònu construction for both sensory modalities, but modifies the clause referring to feeling with the adverbial phrase ní kiṣudzyaɛ ṣu ‘on the skin’.

In all responses, the part of the sentence that refers to hearing was expressed using the verb nu with or without the postposition ònu (depending on the object of the verb). For the part of the sentence that refers to feeling, several ad hoc strategies were used, such as the phrase ‘on the skin’ as in (12), a paraphrase such as ‘it touched me’ or ‘it moved me’, the verb mɔ̀ ‘see’ (see also Section 5.2), or the English verb feel. This was independent of the order in which the two meanings occurred in the prompt, i. e., the same results hold true for sentences such as ‘I did not feel it, but I heard it’. An example of a response with this order can be seen in example (14) in the next section.

These results show that people can disambiguate hearing and feeling when necessary, but there is no conventional way of doing so. In particular, there is no conventional way to unambiguously refer to feeling by touch. There does not seem to be a noun for this sensory modality (analogous to ɔ̀f̣u ‘scent’ and òle ‘taste’) and there is no other clearly preferred strategy for referring to it. The fact that all consultants consistently used nu (+ ònu) to refer to hearing and various different strategies to refer to feeling also provides evidence for hearing as the default interpretation of nu. This will be discussed further in Section 3.3.

2.2 Nu with a subordinate clause

There are two different subordinating constructions with nu. One is constructed with the relativizer gì and is used for direct or immediate perception, where the speaker makes clear that he or she witnessed the event described in the subordinate clause see e. g.,(Dik and Hengeveld 1991; Noonan 1985). An example can be seen in (13).

This construction can be used for both hearing and feeling. It is not used in my data to refer to smelling or tasting, but further research is necessary to establish whether or not this is possible. The construction does not disambiguate between sensory modalities.

The other type of subordinating construction is the use of a complement clause with the complementizer ṣi. In the available data, nu with a complement clause is only used to refer to hearing, but further research is needed to establish whether or not it can refer to other sensory modalities as well. An example can be seen in (14). Like Example (13) above, this is a case of direct perception.

Examples such as (14), where a complement clause refers to direct perception, are not very common in the corpus. More commonly, nu with a complement clause is used with a ‘hearsay’ interpretation. In these cases, the speaker uses nu to indicate that she did not directly witness the event described, but that the information was obtained from another person. An example can be seen in (15).

2.3 How many sensory modalities?

In this section I have described the grammatical properties of the verb nu. At the beginning of this section, I asked the question to what extent the grammatical constructions in which nu occurs disambiguate its meaning. I have shown that they often do: when nu refers to smelling, it usually occurs in a construction with ɔ̀f̣ù ‘scent’ and when it refers to tasting with òle ‘taste’. This means that the linguistic data does not provide evidence for the hypothesis that speakers of Avatime distinguish only two sensory modalities. Even though the perception verbs mɔ̀ and nu make only a two-way distinction, distinctions between other sensory modalities are made by other elements in the grammatical construction. However, the data does not support a perfect mapping onto the traditional five-senses system either, given that the notions of hearing and feeling are not conventionally disambiguated. How this may relate to the conceptualization of sensory perception will be discussed further in Sections 3.3 and 6.

3.1 Perception meanings

The interpretations of nu described in Section 2, ‘hear’, ‘feel’, ‘taste’ and ‘smell’, all refer to non-controlled perception events, in which a perceiver passively experiences a stimulus. The current section discusses to what extent nu can be used for other types of perception events. Verbs referring to non-controlled perception events, also called experience verbs, can be distinguished from activity verbs, which refer to intentionally controlled events (Viberg 1983). Examples of activity verbs are for instance English look at and listen. There is not always a lexical distinction between non-controlled experiences and controlled activities: the English verbs feel, taste and smell can refer to both controlled and non-controlled events and there are many languages in which the distinction is not made for any of the sensory modalities see e. g.,(Aikhenvald and Storch 2013; Evans and Wilkins 2000).

So can Avatime nu be used for controlled perception activities? This depends on the sensory modality referred to. Nu cannot be used for active feel, taste and smell. These are referred to with serial verb constructions. An example can be seen in (16), where the verb gbɔni denotes the activity of getting into contact with something and the visual activity verb di ‘look’ adds the semantics of controlled perception.

The only controlled perception interpretation that nu can have is auditory (‘listen’). An example can be seen in (17).

It is not always straightforward whether to translate an instance of nu as ‘hear’ or ‘listen’. Sometimes, the context will force a particular translation, i. e., ‘listen’ when imperative, as in (17). Most of the time, whether or not a hearing event is controlled or non-controlled simply seems to be left unspecified (for a similar observation, see Aikhenvald and Storch [2013: 19–20]).

If people need to unambiguously refer to a controlled auditory event, they can use a more specific construction: the phrase tã kutu, literally ‘eat/chew ear’. An example, in which the expression is shortened to tã tu, can be seen in (18). The phrase tã (ku)tu is common in the questionnaire translations, which contain context in which the specific reference to controlled perception is made relevant. In the corpus of spontaneous discourse, controlled events are more often referred to with nu and tã (ku)tu is rare.

Another distinction that can be made within perception verbs is between verbs that take the experiencer as their subject and verbs that focus on the source of perception and some quality ascribed to this source. The latter have been called copulative verbs or source-based verbs (Viberg 1983). Examples are English look as in it looks beautiful and sound as in it sounds loud. The Avatime verb nu is not used on its own as a source-based verb, but it occurs in a source-based construction that is used to refer to positively-valued perception events. This construction looks like a serial verb construction with a first verb pe and as the second verb either nu or the verb di ‘look’. The first verb pe, however, occurs only in this construction and is never used as a verb on its own. Examples can be seen in (19) and (20).

The pe + nu combination is only used for hearing and not for feeling, tasting or smelling.

In this section we have seen that nu can be used on its own for controlled perception and as part of a construction for positively-valued source-based perception. In both cases it is only used to refer to hearing and not to touch, taste or smell. This indicates that hearing is more central to the meaning of nu than the other sensory modalities. I come back to this in Section 3.3.

3.2 Non-perception meanings

There are three non-perception interpretations of nu that occur in the corpus. The first is ‘heed’ or ‘obey’. An example can be seen in (21), where the person spoken to likely did perceive what the parents said (otherwise she would not have been able to refuse), but she did not do as they told her.

The second type of non-perception meaning that nu may have is ‘understand/know a language’, as in (22). In this case, the object of nu is either the word ṣìyà ‘language’ or a language name.

Thirdly, nu can also mean ‘understand’ more generally. In this case its object is usually followed by a postposition. This can be either ese ‘under’ or mὲ ‘inside’. An example of the latter can be seen in (23). There are also a few cases where nu is used without any postposition but is still interpreted as ‘understand’. An example can be seen in (24).

All three non-perceptual interpretations of nu are related to the sensory modality of hearing. They are very similar to meaning extensions attested for verbs of auditory perception crosslinguistically (Evans and Wilkins 2000; Vanhove 2008) and in line with Sweetser’s (1990) prediction that hearing verbs will extend their meaning to notions related to linguistic communication (if we assume that a more general notion of understanding may evolve out of the more specific notion of understanding spoken language mentioned by Sweetser).

Perhaps unexpectedly, the verb nu does not have any non-perception interpretations that are related to sensory modalities other than hearing. As discussed in Section 1.1, the sensory modalities of touch and taste are often linked to subjective notions of cognition. This is especially true for verbs like English ‘feel’, which are more frequently used for mental states, emotions or general physical well-being than they are for feeling by touch (Sweetser 1990; Viberg 2008). Avatime nu, despite its use for feeling by touch, does not have any of these interpretations. To talk about emotions, various other constructions can be used, one of which uses the verb mɔ̀ ‘see’, as discussed in Section 4.2. General wellbeing is expressed with a copula construction as in (25).

The fact that the non-perception interpretations of nu are related to the auditory sensory modality, together with the fact that nu does not have the typical non-perception interpretations of ‘feel’ verbs, provides further evidence for a basic auditory interpretation of nu.

3.3 Nu as both general and polysemous

This section addresses two related questions. First, is the verb nu polysemous with separate senses referring to separate sensory modalities, or can it be considered as having a single general perception meaning (i. e., ‘perceive non-visually’). Second, is there one sense or interpretation of the verb that is more central than the others?

Starting with the second question, there is some evidence that shows that the verb nu has a central auditory interpretation. First, the auditory interpretation is the default when people hear the verb nu in a sentence without any context. Second, as was shown in Section 2.1, the auditory interpretation is also treated as default when people are asked to translate sentences that contrast hearing and feeling. Finally, as was shown in Section 3.2, all non-perception meanings of nu are related to auditory perception. These pieces of evidence show that nu is at its core a verb of hearing and the other interpretations are less central.^[8]

The hearing interpretation is not only central to the meaning of the verb, but is also best analyzed as a separate sense. This is shown by applying the so-called definitional test see e. g.,(Riemer 2005). The definitional test assumes that if a word has a general meaning, there should be a definition that includes all the interpretations of the word and at the same time excludes interpretations that the word does not have. At first sight, it seems that there is such a single definition for all perception interpretations of nu, which would be something like ‘perceive non-visually’. However, if we study the semantics of nu in more detail, there is a problem with this definition: the auditory interpretation of nu can occur in more contexts than the ‘smell’, ‘taste’ and ‘feel’ interpretations. This was shown in Section 3.1. Firstly, nu can refer to controlled auditory perception (‘listen’) but not to controlled olfactory, gustatory or tactile perception. Secondly, when nu is used in the pe + nu construction, it refers to source-based, positively-valued auditory perception events (‘it sounds good’) and cannot refer to sensory modalities other than hearing. This means that a definition of the perception meaning of nu as ‘non-controlled non-visual perception’ is too narrow: it does not include the controlled and source-based auditory interpretations. On the other hand, if nu is underspecified for the type of perception event construal (controlled or non-controlled, experiencer-based or source-based), then we would expect it to refer to controlled feeling, tasting and smelling and there is no explanation for why it only refers to audition in the pe + nu construction. Therefore, the auditory perception interpretation is best analyzed as a separate sense. The broad range of applications of this sense compared to the other sensory modalities is also additional evidence for its status as the central meaning of nu.

What about the other perception interpretations? The fact that ‘hear/listen’ is analyzed as a separate sense, does not necessarily mean that ‘smell’, ‘taste’ and ‘feel’ should also each be separate senses of the verb. They might still be collapsed under a more general ‘non-visual perception’ sense. The definitional test shows that such a general interpretation is possible: a definition as ‘controlled non-visual perception’ covers all three interpretations and does not include interpretations that the verb does not have.^[9] Additional evidence for such a general perception sense comes from the constructions in which the ‘smell’ and ‘taste’ interpretations arise. If the interpretation of a word in a particular construction can be predicted based on the other elements in the construction, this shows that a general meaning is possible (cf. e. g., Croft’s [1998] pragmatic model and Tyler and Evans’s [2001] criterion of context-independent occurrences). As was shown in Section 2.1, the ‘smell’ and ‘taste’ interpretations almost always occur when nu is used with the nouns ɔ̀f̣ù ‘scent’ and òle ‘taste’. Given that the information about the specific sensory modality that nu refers to is present in other elements of the construction, separate lexical entries for nu ‘taste’ and nu ‘smell’ are not necessary.

What about ‘feel’? The ‘feel’ interpretation does not arise from the combination of nu with other lexical items and only becomes apparent through context. This could be taken as an indication that ‘feel’ is a separate sense of the verb. However, it could also be the case that the concept of ‘feeling by touch’ is not lexicalized in Avatime at all and that the diverse sensations that are usually considered to belong to the sensory modality of touch are simply talked about as instances of (non-visual) perception more generally. The latter seems to be a more attractive analysis than the former, given that the sensory modality of touch is notoriously difficult to define and includes many sensations that do not appear to have anything in common, such as perceiving texture, weight, temperature and pain for discussion of this issue, see e. g.,(Fulkerson 2015; Macpherson 2011; Sorabji 1971; de Vignemont and Massin 2015). The only proposed definition of touch that includes all these different types of perception is one that defines it as perception with the whole body, or more precisely, perception without a localized sense organ (Sorabji 1971). Unlike vision, hearing, taste and smell, touch is not associated with a particular organ. One might think of the skin, but we do not need the skin to feel touch (e. g., when we touch our teeth), and some sensations, such as weight, are not perceived with the skin at all. This definition basically considers touch as a catch-all category: all types of perception that do not involve the other sense organs. Given that we already had a reason to suggest a sense of nu that includes all non-visual types of perception, it seems unnecessary to consider an additional sense for sensations that are also not hearing, smell or taste.

The evidence presented so far for a general ‘non-visual perception’ sense of nu needs a disclaimer. As has been argued many times, there does not seem to be a single principled way to determine if different semantic interpretations of a word should be regarded as separate senses or as contextually derived meanings see e. g.,(Geeraerts 1993; Riemer 2005; Tuggy 1993). In fact, linguistic evidence alone may never be enough to decide about the mental representations of words (Croft 1998; Sandra 1998). Given that mental representations can be redundant, evidence for a general meaning can never exclude the possibility of polysemy: even if the meaning of a word is always predictable from the construction, it may still be specified in the lexicon as well. This does not pose a big problem for the current investigation. My main aim of presenting the evidence for a general ‘non-visual perception’ sense (in addition to the specific auditory sense) is to show that it is not necessary to propose separate senses referring to tasting, smelling and feeling and that it is possible to account for these interpretations with a general sense. This allows for an account that does not rely on the “Western” five-senses model. To what extent the semantic distinctions proposed in this section map onto the senses that are culturally distinguished and considered important in Avatime will need further anthropological and psycholinguistic investigation.

Taking stock of the evidence provided in this section, a likely analysis is as follows. The verb nu has a central sense referring to auditory perception, which can be either controlled or non-controlled and which can figure in a construction with pe to refer to source-based positively valued perception. In addition it has a more general perception sense that refers to controlled non-visual perception. To talk about tasting and smelling, the verb is used in this second sense and the sensory modality can be specified further by using a construction with a noun referring to the sensory modality. The sensory modality of touch can be considered as a notion that is not part of the semantics underlying the Avatime verb nu and Avatime can be considered not to have a verb or conventionalized construction that specifically refers to non-controlled perception through touch.

4.1 Mɔ̀ as a vision verb

The verb mɔ̀ can occur either with a noun phrase object or with a subordinate clause. Example (26) shows mɔ̀ with a noun phrase object. Unlike nu, mɔ̀ can take any noun phrase as its object.

Like the verb nu (see Section 2.2), mɔ̀ can occur with two types of subordinate clause. The first type has the form of a relative clause and is used for direct or immediate perception, indicating that the perceiver has witnessed the event described in the subordinate clause. An example is shown in (27).

The second and more common type is a complement clause with the complementizer ṣì. This construction can also be used for direct perception, as in (28), but more frequently indicates indirect perception, where the perceiver infers the information in the complement clause, based on visual evidence. The latter case is shown in (29), where the man did not actually see anybody take anything from his farm, but infers that this has happened based on the visual evidence that some of the produce of his farm is gone.

In the remainder of this section, I discuss the interpretations of mɔ̀ outside the domain of perception. These are related to the syntactic construction that the verb occurs in. Some non-perception meanings only occur when mɔ̀ is used with a noun-phrase object and others occur mostly when it is used with a subordinate clause. I discuss these two types of interpretations in turn.

4.2 Mɔ̀ with a noun-phrase object

When used with a noun-phrase object, mɔ̀ can have three types of non-perception interpretation. The first is in the domain of social interaction, with mɔ̀ meaning ‘meet’ or ‘visit’. This is a common interpretation of vision verbs crosslinguistically see e. g.,(Aikhenvald and Storch 2013; Baker 1999; Evans and Wilkins 2000; San Roque et al. 2018). An example can be seen in (30), where it is clear that the speaker does not only want to visually perceive the chief but wants to consult with him.

The second type of non-perception interpretation that is possible when mɔ̀ occurs with a noun phrase object is in the domain of possession. This interpretation is crosslinguistically unusual, but it seems to be relatively common among languages spoken in parts of Africa (Aikhenvald and Storch 2013; Ameka 2013). The verb mɔ̀ can mean ‘find’, ‘get’ or ‘have’. Note that Avatime does not have a verb like English have and using mɔ̀ is one of two conventional ways to talk about possession.^[10] Examples can be seen in (31), which is a case of more literal ownership, and (32), where the concept of possession is extended to abstract properties of people.

The final use of mɔ̀ with a noun phrase object is to express emotions. In these cases, the object of the verb is a noun phrase referring to an emotion, such as ̣ìsàmi ‘happiness’, ḳùṣuyeyì ‘anger’ or ḳùŋẉùza ‘greed/jealousy’. An example can be seen in (33).

The use of vision verbs to talk about emotions seems to be crosslinguistically rare or at least underexplored, but see Section 4.4 for more discussion. From the embodiment perspective proposed by Sweetser (1990), the connection of vision to emotion is unexpected, because vision is claimed to be related to the intellect and not to subjective experiences. In the next section I show more examples of the connection between vision and subjective aspects of cognition in Avatime.

4.3 Mɔ̀ with a subordinate clause

When mɔ̀ occurs with a subordinate clause, it can refer to both objective knowledge and to the more subjective notions of belief and personal opinion.

The knowledge interpretation is common for vision verbs crosslinguistically cf. e. g.,(Baker 1999; Evans and Wilkins 2000; Johnson 1999; Sweetser 1990) and seems a natural extension from indirect perception cases like (29), where knowledge is acquired through inference based on vision. An example of mɔ̀ referring to the acquisition of knowledge without this being based on vision (only) can be seen in (34). Here, the realization of the man that he is in trouble is not just based on his visual perception of the situation he is in, but also on his interaction with the crocodile. Another example is (35), where visual perception plays no role in the insight that the history should be written down.

The subjective interpretations of vision verbs have been less frequently explored in the literature (but see Section 4.4 for more discussion). An example of mɔ̀ referring to a personal point of view is line 2 of (36), where speaker B uses mɔ̀ to give her opinion about what she thinks is happening in the picture they are describing.

Another example is (37). Here, in line 5, speaker B uses mɔ̀ to indicate that it is her personal belief that the reason why the woman under discussion washed her t-shirt is that she wore it. She has not actually seen the woman wear the t-shirt. The latter is clear from line 2, in which she says she does not understand why the woman should wash her t-shirt.

Another example that shows the use of mɔ̀ to indicate a personal opinion is (38). Here, mɔ̀ is used to indicate that the amount of money to donate is up to each person’s individual choice. Note that the most natural English translation uses the perception verb feel.

This use of mɔ̀ for subjective, personal opinions contradicts Sweetser’s (1990) prediction that vision verbs only refer to objective knowledge. As was mentioned in Section 1.1, Sweetser argues that the association between vision and objective knowledge has its roots in the nature of vision itself. Because we see things from a distance, we also keep a personal distance to the knowledge gained through vision. In addition, we assume that what other people see is the same as what we see and therefore we consider the knowledge gained through vision to be objective. The Avatime data show that this is not universally true and that vision may also be conceptualized as a more private and personal sensory modality.

4.4 Seeing, hearing and cognition

As we have seen in the previous section, the verb mɔ̀ ‘see’ is used with reference to objective knowledge, as well as personal opinions and beliefs. The latter, subjective, conceptualization of vision is also reflected in the use of mɔ̀ to express emotions, as described in Section 4.2. It also aligns with the use of mɔ̀ for possession, also discussed in Section 4.2. (Ameka 2013: 238), writing about the use of vision verbs for possession in the Ghanaian Kwa languages Likpe and Ewe, comments that this “reflects the idea that what is in one’s perceptual domain belongs to them”. The idea that what I see belongs to me follows more naturally from a view of vision as private and personal than from a view of vision as distant and objective.

The use of vision verbs for subjective cognitive notions has hardly been discussed in the literature, but is not unique to Avatime.^[11] Vision verbs can be used for emotions in Kambaata and several other Ethiopian languages (Treis 2010) as well as in Luganda (Thanassoula 2012) and Ewe (Ameka 1990). Using the vision verb for talking about one’s personal opinion can also be done in English, using phrases such as the way I see it or as I see it. A difference with Avatime is that English uses special constructions for this purpose with elements such as way, as or like, which also contribute to the subjective reading. In Avatime, the same basic complement construction is used for vision as well as for both objective and subjective cognition. Another subjective notion that is expressed with the vision verb in several languages is what San Roque et al. (2018) call subjective co-identification. This notion is also expressed with a specific construction. In English this is the construction see X as Y, for instance I never saw it as a radish (San Roque et al. 2018: 383). Finally, a quick search through RefLex, an online collection of dictionaries and word lists of African languages (Segerer and Flavier 2011–2018), also yields several cases of what seem to be subjective interpretations of vision verbs. For example, in Mano (Mande) the verb g[epsilon] glossed as ‘voir’ (‘see’) has an additional meaning ‘penser’ (‘think’), which could indicate a subjective interpretation, as well as ‘considérer (comme)’, (‘consider as’) which seems to be an instance of subjective co-identification (Khachaturyan 2012). Another example is Pular (Atlantic), in which the verb yi’ugol glossed as ‘voir’ has the additional gloss ‘penser’ but also ‘être de l’avis que’ (‘be of the opinion that’) (Bah 2009). More in depth research on perception verbs in individual languages is needed to understand the variety of subjective notions that can be expressed by vision verbs, but it is clear that an account of vision as inherently linked to objective cognitive notions only cannot be upheld in the light of crosslinguistic data.

The non-vision verb nu can refer to obeying, knowing a language and understanding (see Section 3.2). This makes nu very different from general non-visual perception verbs in several other languages. French, Spanish and Italian also have general non-vision verbs, but these have various non-perception interpretations in the domains of knowledge, personal opinion and emotion (Enghels and Jansegers 2013). For instance, the Spanish verb sentir conveys “subjectivity and epistemic modality: the perceiver’s feelings, emotions, attitude, judgment, thoughts and viewpoints” (Guerrero and Cruz Domí nguez 2017: 369). The general verb hulira in Lussese (Bantu) has a wide range of non-perception uses, related to social interaction, cognition and emotion (Thanassoula 2012; Thanassoula 2013). In these languages, the broad range of perception interpretations corresponds to a broad range of non-perception interpretations. This is not the case in Avatime, where the general non-visual perception verb nu has a broad range of perception interpretations, but only a few meanings outside the domain of perception.

Going back to the Avatime vision verb mɔ̀, it is striking that its interpretations outside the domain of perception are very similar to those of the general perception or non-vision verbs in Romance languages and Lussese mentioned above. The interpretation of mɔ̀ in the domain of emotions also shows a similarity to verbs for feeling by touch in other languages, such as English ‘feel’. The verb mɔ̀ thus has many of the non-perception interpretations that might have been expected for the verb nu. This provides evidence for the idea that there is a potentially universal link between perception and cognition, but that the connections between particular sensory modalities and particular cognitive meanings are more language specific see also(Evans and Wilkins 2000; Ibarretxe-Antuñano 2008; San Roque et al. 2018).

5.1 Nu in discourse

As was mentioned in Section 1.2, the corpus contains 839 cases of mɔ̀ and 223 cases of nu. The difference in number of cases is in line with the finding by San Roque et al. (2015) that people across languages talk more about vision than about the other sensory modalities. For the non-visual sensory modalities San Roque et al. find variation in which is mentioned most frequently, but in most languages hearing comes in second place and is well-ahead of touch, taste and smell. Avatime, part of San Roque et al’s sample, is one of these languages. The data used for the current paper, taken from a larger corpus, confirm this picture. Out of the 165 cases of nu that refer to perception events (the other 58 cases are non-perception meanings), the great majority, 152, refer to auditory perception. There are 11 cases of nu referring to smell, 2 cases referring to touch and no cases referring to taste.^[12]

The 11 cases of nu referring to smell are all marked as such by the use of the noun ɔ̀f̣ù ‘scent’, as in example (39). This shows that even though ɔ̀f̣ù is not obligatory when talking about smell (see Section 2.1), there is a clear preference to use it.

Of the two cases of nu that refer to touch, one occurs with the object kuyeyeyi ‘pain’, which immediately makes the sensory modality clear. In the other case, which is shown in (40), the sensory modality is ambiguous without context. This sentence was originally translated by a language consultant as ‘you will hear the sound kpu’. However, somewhat later in the discourse it becomes clear that the speaker is talking about the experience of a pregnant woman when the baby kicks or moves inside the womb, so a more appropriate translation is ‘feel’. Note that this supports the analysis of hearing as the central sense of nu (Section 3.3): the translator’s first thought is hearing, but when context made clear this was not possible, a different interpretation was found.

When nu refers to hearing, it most frequently occurs without any specification of the intended sensory modality and it is the context that makes the interpretation clear. An example can be seen in (41). Here, the use of tã tu ‘listen’ in line 2 makes clear that nu in line 3 should be interpreted as referring to auditory sensory modality.

There are five cases in which the interpretation as auditory is also clear without the wider context because of the object of nu. One of these can be seen in (42), where the object is the noun kunugu ‘talking’. There are also 22 cases of nu with a subordinate clause, in which it is usually the content of the subordinate clause itself that makes the sensory modality clear, as in (43).

There are no cases in the corpus where the general meaning of nu is used to purposefully refer to perception with an unspecified sensory modality or to perception with multiple modalities at once. This shows that the broad meaning of nu is not related to a more general tendency to talk about perception in a non-specific way.

It is notable that this is not even the case in contexts in which non-visual perception is particularly relevant. This can be seen in an interview with a blind woman about how she lives her daily life. We might expect her to use nu to talk about her sensory experiences, given that these are all non-visual. However, she does not do this at all in the interview and instead uses the verb mɔ̀ ‘see’. An example can be seen in (44).

Given that the speaker in example (44) must be using her non-visual sensory modalities to sense that she has passed the junction (probably a combination of hearing and touch), it seems unexpected that she does not use the verb nu. In the next section I will discuss some related examples and propose an explanation.

5.2 Mɔ̀ for non-visual perception

In the previous section, an example was shown of mɔ̀ used to refer to a non-visual perception event. There are several other cases of this in the corpus. Two of these can be seen in (45) and (46).

In these cases of mɔ̀ used for non-visual perception, mɔ̀ always occurs with a subordinate clause. This is the same grammatical construction in which mɔ̀ can be interpreted as referring to the domain of cognition (Section 4.3). This connection to cognition seems to be what drives the use in non-visual perception contexts. In example (45), the speaker uses mɔ̀ to refer to knowledge that she has obtained through non-visual perception, in this case touch. However, her main message is the fact that she has acquired the knowledge that the weeds have grown tall. Given that mɔ̀ can be used to talk about the acquisition of knowledge more generally, it can also be used in cases where this acquisition of knowledge is based on non-visual perception. In (46), mɔ̀ is used to describe a tasting event. However, the main message is that the speaker approves of the salt level. The verb mɔ̀ is used here because of its more general connection to personal opinion.

In the examples above, the subordinate clauses do not describe perceptual experiences directly, but rather conclusions drawn from perceptual experiences. For the speakers who produced these utterances, the main point was not to describe a perceptual experience, but to convey a particular realization or belief. One could therefore argue that they are simply cases of the cognitive use of mɔ̀ and have little to do with perception. However, there are also cases of subordinate clauses used with mɔ̀ that directly describe non-visual perceptual experiences. Most of these cases refer to the sense of touch, as exemplified by (47) and (48) below. There is one case in the corpus of mɔ̀ being used with an auditory experience, shown in (49).

In examples (47) – (49), the speakers clearly focus on the description of a perceptual experience rather than on the cognitive implications of this experience, but the verb mɔ̀ is still used. Uses such as these may indicate a further stage of semantic change of the verb mɔ̀, where cognitive interpretations based on non-visual evidence pave the way for the interpretation of mɔ̀ as referring to non-visual perception more directly (for a similar suggestion based on a varied sample of languages, see San Roque et al. [2018]). However, a question that arises is why mɔ̀ is used at all in these cases, given that the non-vision verb nu is also available. To understand this, we need to consider how nu is used with complement clauses.

As mentioned in Section 2.2, nu with a complement clause often indicates that the information in the complement clause was obtained through hearsay (see Example (15)). This means that if speakers were to use nu instead of mɔ̀ for examples such as (47) – (49) above, there would be a potential for misunderstanding: the addressees might interpret the information in the subordinate clause as obtained from a third party rather than through personal experience. The verb mɔ̀, because of its connection to personal opinion and belief, is then closer to the meaning the speaker wants to convey. This also explains why nu is not used in the blind woman’s narrative for cases like (44) and (45). If she had used nu, she might have given the wrong impression that she was talking about information that she obtained through hearsay. Her use of the verb mɔ̀ makes it clear that she is talking about her personal experience.

The fact that most cases in which mɔ̀ is used for non-visual experiences refer to touch and there is only one case referring to hearing is most likely related to the default hearing interpretation of nu. Despite the potential misunderstanding as hearsay instead of direct auditory perception, people still prefer to use nu for hearing because it is so closely associated with this sensory modality. For touch, the sensory modality will have to be derived from context, irrespective of whether nu or mɔ̀ is used, leading to a preference for mɔ̀ in many cases.