Semantic maps and temperature: Capturing the lexicon-grammar interface across languages

Abstract: Traditionally, lexical typology has to a large extent been interested in lexical categorization of various cognitive domains (e. g., colour, perception, body), i. e., in how these are cut up by the most important words in a language, and in lexical motivation, or formal relatedness, i. e., in whether words for certain concepts are completely unrelated or related to others via polysemy or derivation (e. g., intransitive vs. transitive verbs, words for ‘day’ and ‘sun’, etc.).Grammatical behavior of words andmorphosyntactic patterns as encoding meanings traditionally belong to grammatical typology. In this paper, I consider the domain of temperature and show how the close interaction between lexicon and grammar in the encoding of the temperature domain across languages calls for an integrated lexico-grammatical approach to these phenomena. As auseful tool for such an enterprise I suggest an elaborated semantic map comprising three layers – the layer specifying the words with the information on their mutual formal relations (i. e., whether they are identical, completely unrelated or related via derivation or inflection), their morphosyntactic properties (e. g., their part-of-speech affiliation, inflectional potential, etc.), and the constructions they occur in.


The temperature domain and frames of temperature evaluation
Temperature is usually included among property concepts, e. g., as a subcategory of the class of physical properties in Dixon's (2004) influential list of concepts that are often lexicalized as adjectives. However, temperature concepts have a remarkable status among property concepts in applying to very different states of affairs. Now, dimension, evaluation or age can of course also pertain to wildly different entities and situations. However, what probably singles out temperature concepts is the fact that they are regularly found in three (or even four) fundamentally different kinds of situations (ex. (1)), which will be called frames, subdomains, or kinds of temperature evaluation (Koptjevskaja-Tamm 2015). The notion of a frame, as used here, corresponds to the notion of semantic frames introduced by Fillmore (1977;1985) and utilized in FrameNet (http://framenet.icsi. berkeley.edu/) as a description of a type of event, relation, or entity and the participants in it. 1 (1) Frames of temperature evaluation a. tactile temperature, or "touch-temperature": The stones are hot. b. personal-feeling temperature: I am hot. c. ambient temperature: It is hot here; The summer was hot.
First of all, there are uses like The stones are hot/cold or hot/cold stones, in which temperature concepts describe properties or states of particular concrete entities, carriers of properties, and align with property concepts in general ('dark', 'big', 'sharp', etc.). These uses belong to the tactile temperature frame. Then there are examples like I am/feel hot/cold, which describe a very subjective experience of personal-feeling temperature. These are bodily sensations involving sentient, usually human experiencers, on a par with others, such as feeling hungry/thirsty, having pain etc. (Verhoeven 2007: 42-43). A person can feel hot or cold due to various reasons, which makes bodily sensations different from physical properties and states.
Finally, we find temperature expressions in examples like It is hot/cold here or a hot/cold day, which describe temperatures related to certain circumstances -these are ambient temperatures. In contrast to both tactile and personalfeeling temperatures, predications about ambient temperatures do not have to involve participants.
However, ambient temperatures may also be construed as properties of particular entities, both concrete and abstract -places (indoors, such as houses and saunas, or outdoors, such as countries), sources and conductors of heat (sun, fire, air, wind), weather and climate, and time periods (days, months, etc.). The contrast between the English sentences It was cold in March/in the house and (2) Wolof: temperature predication, constructions with quality verbs (Perrin 2015) a. Tactile temperature: Kafe bii dafa tàng. coffee this 3sg.vf be.warm/be.hot 'This coffee is warm/hot.' (Jean Léopold Diouf, p. c.) b. Quasi-referential ambient: Néeg bii dafa tàng.
room this 3sg.vf be.warm/be.hot 'This room is hot.' (Perrin 2015: 165) c. Non-referential ambient: Dafa tàng ba ëpp. 3sg.vf be.warm/be.hot until be.excessive 'It's excessively hot.' (Perrin 2015: 166) d. Personal-feeling: Dama tàng! 1sg.vf be.warm/be.hot 'I feel hot!' (Perrin 2015: 166) Languages like English or Wolof, which use the same central temperature lexemes in (almost) the same constructions for all the frames of temperature evaluation, are opposed to others, in which the frame-related distinctions have clear manifestations. The Indo-Aryan language Palula spoken in Pakistan (phal1254) is a case in point (for details on the temperature term system, see Liljegren and Haider 2015). Palula makes a basic distinction between 'cold' and 'hot/warm'. However, for each of these values there is a lexical split between two different lexemes -a temperature adjective, involved in predications about tactile and quasireferential ambient temperatures, and a temperature noun, involved in predications about non-referential ambient and personal-feeling temperatures. Predications about tactile and quasi-referential ambient temperatures follow the standard morphosyntactic pattern of adjectival predication: in ex. (3), the tactile (and quasi-referential ambient) 'hot' is expressed by the adjective táatu/téeti in the nominative case, with or without a copula verb and agreeing with the subject in gender and number. The semantic difference between tactile and quasi-referential temperatures in Palula is therefore not reflected in either lexical or constructional differences. In contrast to this, predications But ex. (2c) can also be interpreted as referring to tactile, quasi-referential and personal-feeling temperatures when the reference of the subject is clear from the context. about non-referential ambient and personal-feeling temperatures involve the unrelated noun húluk, but there is a radical constructional split between the two uses. Non-referential ambient temperature is typically construed as an entity whose existence or coming into existence is stated, often in relation to a particular location or time (Liljegren and Haider 2015: 454), cf. ex. (3). Such constructions appear with or without an overt copula or with phrasal verbs like 'become'. Finally, in constructions for personal-feeling temperatures, as in ex. (3d), the temperature noun is the subject of verbs like 'falling' and 'rising' and the Experiencer takes the non-nominative case, otherwise used for the expression of location or goal (Liljegren and Haider 2015: 455). Across the South Asian languages cases akin to these are well known as Dative Subject constructions, or, as experiencer-as-goal constructions, following Bickel (2004: 84-89).
(3) Palula (Liljegren and Haider 2015) a. Tactile (Liljegren and Haider 2015: 452) c. Non-referential ambient: páar bıíḍ-u húluk de. last.year much-nom.m.sg heat[m] be:pst 'It was very hot last year.' (lit. 'Last year there was much heat.') (Liljegren and Haider 2015: 458) d. Personal-feeling: asaám húluk dít-u de. 1pl.acc heat[m] fall.pfv-msg be:pst 'We were feeling hot.' (lit. 'Heat had fallen on us.') (Liljegren and Haider 2015: 456) The examples quoted above raise two different but interrelated questions that will be in the focus of the present paper: 1. How do the lexicon and grammar interact with each other in the expression of the distinctions among the different frames of temperature evaluation across languages? 2. How can this interaction be represented?
The languages considered in the paper come from the large-scale language sample (currently about 70 languages) for which the data have been systematically collected by means of questionnaires and dictionaries in close collaboration with experts and native speakers (cf. Koptjevskaja-Tamm 2015 for the details on the methodology). For the purpose of this study, I have selected 27 languages to illustrate some of the interesting cases without any further requirements on how well they represent the world's languages. The languages and the sources of the data are listed in Table 4 (presented at the end of the paper).
I will focus on the central temperature terms, to the exclusion of the more marginal ones such as scorching or tepid that have a very restricted application. Languages vary as to how many temperature values they distinguish, i. e., whether they only have a binary distinction between 'hot/warm' vs. 'cold/cool' or have more elaborated systems (i. e., 'hot' vs. 'warm' vs. 'cold/cool', or 'hot' vs. 'warm' vs. 'cool' vs. 'cold'). This will not be discussed in the present study but will be visible in the tables and semantic maps for the different languages (cf. Koptjevskaja-Tamm 2015 for the argumentation behind these distinctions).

The challenge of creating a semantic map for the temperature domain
In my earlier work (Koptjevskaja-Tamm 2011;2015: 17), I have suggested that the cross-linguistically recurrent patterns in how languages distinguish among the frames of temperature evaluation by lexical means can be captured in a semantic map like the one in Figure 1. Figure 2 shows how five different languages carve up this semantic space lexically for either 'cold' (Modern Eastern Armenian, Kamang and Japanese) or 'hot/warm' (Palula and Wolof). 5 As evidenced by Figure 2, none of the five different language-specific configurations violate the connectivity hypothesis, according to which "any relevant language-specific and construction-specific category should map onto a connected region in conceptual space" (Croft 2001: 96). The main generalization is that personal-feeling temperatures are often singled out by languages, whereas the linguistic encoding of ambient temperature may share properties with those of either tactile or personal-feeling temperature. Semantic maps like those in Figure 1 and Figure 2 look like regular semantic maps for the purposes of lexical typology, which show how a certain cognitive domain (e. g., colour, perception, body) is carved up by (central) words, 5 Importantly, linguistic temperature systems are often heterogeneous and may consist of several subsystems, with the consequence that the different temperature values (like 'hot' and 'cold') even within one and the same languages may sometimes differ in their lexical and constructional distinctions.
ideally belonging to the same word class (cf. Koptjevskaja-Tamm et al. 2015, Georgakopoulos et al. 2016, and Rakhilina et al. 2022 on semantic maps in lexical typology). Translated into François' (2008) terminology, such maps reflect patterns of colexification by visualising which concepts are associated with one and the same word. François (2008) suggests further to distinguish between strict colexification, defined on the basis of identity of forms in synchrony, and loose colexification, which covers relatedness of forms encoding two concepts from a diachronic point of view as well as cases of partial identity of forms, for instance, in derivation or compounding. For instance, the two 'hot/warm' words in Palula (táatu/téeti vs. húluk are not formally related at all, while the 'cold' adjective and noun (šidáalu vs. šid) are clearly related, at least diachronically. Thus, as shown in Figure 3, Palula strictly colexifies the concepts 'tactile cold' and 'quasi-referential cold' by šidáalu, and 'non-referential ambient cold' and 'personal-feeling cold' by šid, but loosely colexifies 'quasi-referential cold' and 'non-referential cold' by šidáalu vs. šid. In other words, even the non-identical temperature words used for different  functions can differ in their formal relatedness, or motivational patterns (Koch 2001), i. e., whether they are completely unrelated or are related via derivation (synchronically or diachronically).
The problem is, however, that for the temperature domain the lexical and grammatical/constructional distinctions are often inseparable from each other. As we have seen in ex. (3) in Section 1, the two 'hot/warm' words in Palula differ in their word-class affiliation (adjective vs. noun) and are used in very different constructions, and the same goes for the two 'cold' words in Figure 3. Likewise, the personal-feeling 'cold' in both Modern Eastern Armenian and Kamang is a verb, whereas the other 'cold' words are adjectives. Figure 4 shows how the same semantic space as in Figure 2 (of 'cold' or 'warm/hot') is organized in the same five languages, but this time in terms of different word classes. The resulting configurations, again in accordance with the connectivity requirement, are fewer than those in Figure 2 with the result that Modern Eastern Armenian and Kamang share the same pattern, and the same goes for Japanese and Wolof.
Finally, when constructional similarities are taken into account, the corresponding semantic maps will again differ and unveil still other cross-linguistic similarities and dissimilarities. For instance, as shown in ex. (3) in Section 1, Palula has three radically different constructions for the different frames of temperature evaluation, whereas in Modern East Armenian (not exemplified here), the non-referential ambient construction is closely related to the one used for tactile and quasi-referential ambient predications.
To sum up, already these examples testify to an intricate picture of how different linguistic means interact with each other in carving up the same semantic domain and expressing the relevant categories. But how exactly is this joint linguistic expression of categories carried out across languages? In other words, is there any systematicity in this interaction? Or, to quote Evans (2010), "how do semantic choices made in one subsystem affect those in others?" These questions are in fact quite problematic, seen against the background of the customary division between lexical and grammatical typology. Grammatical behavior of words and morphosyntactic patterns as encoding meanings traditionally belong to grammatical typology, which normally disregards the details related to lexical choice. For lexical typology, on the other hand, lexical choice is of primary interest, but the grammatical behavior of the relevant lexical expressions lies normally outside of its scope. This is all the more unfortunate given that both grammatical and lexical typology share much of their methodology, including ample use of semantic maps for representing their generalizations (cf. Georgakopoulos and Polis 2018 for an excellent overview). 6 However, this problem has been noticed earlier and linguistic literature has several suggestions on how to reconcile description, analysis and representation of both lexical, grammatical and syntactic information cross-linguistically. One promising direction starts from the methodology of semantic maps and enrichens them with constructional information, as this is done in the comparison of the uses of manner demonstratives in Kambaata (Cushitic) and Wan (Mande) in Nikitina and Treis (2020), and in the semantic comparison of the words for 'grain' and their relatives across South Mande in Nikitina (2022).
Another direction starts from the framework elaborated in constructional approaches to language, which reject the idea of compartmentalization in linguistic analysis, including the sharp distinction between lexicon and grammar, and 6 Not surprisingly, this is also the case in most of the contributions to this volume: e. g., Rakhilina et al. (2022), and François (2022) deal exclusively with wholesale content lexemes, without considering their grammatical behaviour and the constructions in which they are used, while Vanhove (2022) is interested in the grammatical markers, without considering lexical semantics. And while Levshina (2022) and Becker and Malchukov (2022) discuss issues on the crossroads between lexicon and grammar (morphology and constructions), they do not spell out how the details of this interaction can be captured and represented. consider every kind of conventionalized pairings of meaning and expression as constructions (Croft 2001;Goldberg 1995;2006). However, most of the research in constructional grammar has so far been very language-specific, with the crosslinguistic research restricted to morphosyntactic patterns (Croft 2001, to appear). Koch (2012) offers a major consistent attempt to apply a constructional approach to the cross-linguistic analysis of joint linguistic expression of a semantic space, in this case location-existence-possession. What makes this "super-domain" particularly interesting and directly comparable to the temperature domain is that its encoding involves an intricate interaction between lexical and morphosyntactical means. For instance, while the verb ter 'to have' in Brazilian Portuguese jointly lexicalizes possession and existence, the latter meaning is encoded by an impersonal construction, as opposed to the normal subject-predicate construction for possession, cf. o rapaz tem um livro 'The boy has a book' vs. tem um livro. These aspects have been to a certain extent downplayed in the rich earlier crosslinguistic research on location-existence-possession, which has been partly split between the studies primarily interested in joint lexicalization and those primarily dealing with morphosyntactic patterns. Koch (2012) sets out to demonstrate how Construction Grammar can be utilized to provide an integral account for the complex lexicon-grammar interaction, or "constructional typology" in a sample of nineteen genetically and areally diverse languages.
My own proposal to integrate the traditions of grammatical and lexical typology, elaborated in the rest of this paper, is reminiscent of the one in Nikitina and Treis (2020) and in Nikitina (2022) in using semantic maps and enrichening them with grammatical and constructional information. On the other hand, it pays more attention to the structural similarities between the constructions (building on the notion of inheritance), much in the spirit of Koch (2012). In what follows I will examine a few cases in which different linguistic means interact in carving up the temperature domain and reflect on how these can be captured by means of elaborated semantic maps.
I will start from the grammatical typology, i. e., from the constructions used for the three kinds of temperature predications, with constructions for personalfeeling temperatures as pivotal for the ensuing discussion. These will be introduced in Section 3. 7 In Sections 4 and 5, I will focus on languages with two particular groups of constructions for personal-feeling temperatures and investigate how these relate to the constructions used for the other frames of temperature evaluation and to the choice of the temperature term involved in them. To account for these relations, I will introduce an elaborated model of semantic maps with three layers -the layer specifying the words with the information on their mutual formal relations (i. e., whether they are identical, completely unrelated or related via derivation or inflection), their morphosyntactic properties (e. g., their part-ofspeech affiliation, inflectional potential, etc.), and the constructions they occur in. Section 6 will conclude the article.

Constructions for temperature predications
As announced in Section 1, one of the main goals of this paper is to try to understand how the lexicon and grammar interact with each other in the expression of the distinctions among the frames of temperature evaluation across languages. A crucial question in this connection is which types of predicative constructions for the different functions (i. e., for the different frames of temperature evaluation) may co-exist within one and the same language, and/or whether these involve the same or different lexical items. This is the question which I want to explore in the rest of this paper.
Predicative constructions of various kinds have been of central concern for grammatical typology. There are several typologies of property predications, where the main focus has been on the extent to which these align with verbal, existential, nominal and locational predications (cf. Hengeveld 1992;Wetzer 1996;Stassen 1997;Pustet 2003). Cross-linguistic research on experiencer constructions, including constructions for bodily sensation, is relatively rich but less systematic. One of the pertinent issues here is to what extent experiencers are encoded as (canonical) agent-like subjects or appear as datives or as patients (e. g., Aikhenvald et al. 2001;Bhaskararao and Subbarao 2004). Reh (1998) suggests a framework for capturing cross-linguistic variation in experiential constructions and applies it in the brief descriptions of a few African languages. Reh's framework is further elaborated in Verhoeven's (2007) detailed study of experiential constructions in Yucatec Maya. Systematic cross-linguistic research on ambient constructions is relatively limited, apart from Eriksen et al. (2010), who propose a general typology for capturing the cross-linguistic variation in weather expressions with the main focus on precipitation constructions, and scattered descriptions and mentions in connection with impersonal constructions (e. g., in Malchukov and Siewierska 2011).
In general, studies focusing on constructions used for property, experiencer and ambient predications are normally restricted to one of these functions and have hardly anything to say about its potential links to the others. This is understandable given that very few concepts can be found in more than one of the three functions. Pustet (2015) is noteworthy in zooming in on the constructions that apply to the lexical expressions coming from the same domain of temperature, but used in different functions. Since constructions for tactile predications are unremarkably uniform in having the entity whose temperature is evaluated as the subject of the intransitive temperature predicate, Pustet focuses on personal-feeling (experiencer) and non-referential ambient temperatures. She suggests typologies for both, but taking each of the functions per se. As a result, the two typologies are neither directly commensurable with each other nor sufficiently fine-grained for providing answer to the question of which types of constructions for the three functions may co-exist within one and the same language, and/or whether these involve the same or different lexical items.
In the following, I will make use of Pustet's observation that personalfeeling temperatures show the greatest cross-linguistic variation in their constructional patterns as compared to both ambient and tactile temperatures. I will ask whether the choice of a particular construction for personal-feeling temperatures has consequences for the choice of constructions for quasireferential and non-referential ambient temperatures and/or for the lexical distinctions between the temperature terms used in the four functions. My point of departure will therefore be a classification of personal-feeling temperature constructions, largely inspired by both Verhoeven (2007) and Pustet (2015), as presented in Table 1.
The two organizing parameters in Table 1 are the valency of the main predicate (intransitive vs. transitive) and the grammatical subject of the construction, or "the goal of orientation" (Verhoeven's term). Personal-feeling constructions can thus be Experiencer-oriented, Expertum-oriented (i. e., having the sensation/feeling, in this case temperature, as the subject), and Non-oriented (i. e., lacking a subject or having an ambient/"dummy" subject) constructions. 8 Relating to the examples quoted so far, personal-feeling predications in both English (ex. (1b)) and Wolof (ex. (2d)) follow the Experiencer-oriented intransitive strategy, with temperature adjectives in English and temperature verbs in Wolof, whereas personal-feeling predications in Palula (ex. (3d)) use the Expertum-oriented intransitive strategy involving temperature nouns, the motion verbs 'rise'/'fall' and the Experiencer as the non-core argument. The German example Mir ist kalt lit. 'To_me is cold' illustrates the Non-oriented intransitive strategy. Table 1: A typology of construction types used for personal-feeling temperatures inspired by Pustet (2015) and Verhoeven (2007: 71-72) (EXPr = Experiencer, TempN = temperature noun). In the next two sections I will focus on two groups of languages: those in which personal-feeling temperature predications involve temperature nouns (as in Palula), and those in which personal-feeling temperature predications follow the Non-oriented intransitive strategy (as in German). I will present an array of cases to show the interplay between constructional and lexical choices and discuss how these constructional and lexical similarities and distinctions can be captured in semantic maps. In the rest of the paper, I will use the abbreviations TACT, AMB_QR, AMB_NR and PF for tactile, quasi-referential ambient, non-referential ambient and personal-feeling temperatures, respectively.

Languages in which constructions for
personal-feeling temperature predications involve temperature nouns Table 1 contains three different construction types pertaining to PF temperature and involving temperature nouns -two Expertum-oriented strategies and the Experiencer-oriented transitive strategy. As already mentioned, Palula ex. (3d) illustrates the intransitive Expertum-oriented strategy. Ex. (4) from Likpɛ (Sɛkpɛlé) (sekp1241), a Kwa language spoken in West Africa, shows the transitive Expertumoriented strategy with the verb 'hold', while ex. (5) from Maltese (malt1254), an Arabic (Semitic) language spoken in Malta, follows the transitive Experienceroriented strategy with the verb 'feel' taking the temperature noun (with or without the definite article) as its object.
(4) transitive expertum-oriented pf: Likpɛ (Sɛkpɛlé) (Ameka 2015: 48) lɛ-yɔ lɛ́mɛ cm-cold hold 1sg 'I am feeling cold' (lit. 'Cold is holding me') (5) transitive experiencer-oriented pf: Maltese (Michel Spagnol p. c.) Qed in-ħoss (is)-sħana prog 1sg.ipfv-feel def-heat 'I am feeling the heat' Table 2 (presented at the end of the paper) displays the constructions and the temperature words used for TACT, AMB_QR, AMB_NR and PF predications in a sample of ten languages in which PF constructions involve temperature nouns. Now, knowing that PF predications in a language are based on temperature nouns, can we draw any further conclusions on the properties of the linguistic expressions involved in the encoding of the temperature domain in the language? There are at least three groups of further questions to ask here -questions about the morphosyntactic properties of the temperature words, their formal relatedness, and the constructions used for the different frames of temperature evaluation and the relations among them. These three groups of questions correspond to the three layers in semantic maps, as will be elaborated in the rest of this paper.
First, there are further questions related to the morphosyntactic properties of the temperature words, such as whether the use of a temperature noun in PF predications has implications for the word-class affiliation of the temperature term(s) involved in predications for the other frames of temperature evaluation.
Clearly, given that TACT and AMB_QR predications are normally based on adjectives or on stative verbs, there will be a radical word-class split between the temperature terms involved in them, on the one hand, and in PF predications, on the other. But what about AMB_NR predications?
The data in Table 2 seem to suggest the following generalization: If a language has a PF predication based on a temperature noun, it also has an AMB_NR predication based on the same noun.
(6) amb_nr predication: Likpɛ (Sɛkpɛlé) (Ameka 2015: 48) lɛ-yɔ kpé cm-cold be.in 'It is cold' The reverse is not true, as shown in ex. (8) from Modern Greek, which uses temperature nouns in the accusative case as the object of an impersonal transitive verb 'make/do' in AMB_NR predications (ex. (8c)), but temperature verbs in PF predications (ex. (8d)) (and temperature adjectives in both TACT and AMB_QR predications, ex. (8a)-(8b)). The semantic map in Figure 5 visualizes the above generalization in terms of the morphosyntactic properties of the temperature term involved in the predications for the different frames of temperature evaluation, whereas Figure 6 shows the corresponding morphosyntactic map for Modern Greek. Second, there are questions pertaining to the choice of the temperature words themselves and their formal relatedness, or motivational patterns, i. e., the extent to which the terms are related to each other in their form, mirroring their conceptual relatedness (Koch 2001). All the ten languages in Table 2 use the same temperature noun in both AMB_NR and PF predications, but show an interest-  ing cross-linguistic variation when it comes to its formal relatedness to the corresponding adjective/stative verb.
Finally, there are cases like Italian 'hot' (caldo) and 'cold' (freddo) and Spanish 'cold' (frío), where the same form is used in all constructions, the only difference being the presence of gender-number agreement in the TACT and AMB_QR uses. There are therefore no clear indications as to the word-class status of the forms used for AMB_NR and PF predications, apart from their uses in typical nominal functions as direct objects to the verbs 'make/do' and 'have', which they share with obvious nouns such as 'day' or 'hunger' in other ambient predications and experiential predications about bodily sensations. Importantly, all the three words have other typically nominal uses (e. g., Luraghi 2015: 338-339 treats caldo and freddo in Italian as both nouns and adjectives). It is therefore not too far-fetched to treat these examples as adjective-to-noun conversion, i. e., a special case of derivation, and loose, rather than strict colexification in François' (2008) terms, and to analyze AMB_NR and PF predications as nominal. Figure 7 elaborates the simple semantic map in Figure 5 based on the morphosyntactic properties of the temperature words involved in the different predications, by including an additional layer representing the formal relatedness between them. Here the 'cold' noun and adjective/stative verb stretch between being completely unrelated (in Bamana) via clearly related by an explicit derivational marker (in Maltese) to being related via conversion (and having basically the same form) (in Italian).

Figure 7:
A two-layered semantic map for temperature predications for 'cold' containing information both on the morphosyntactic properties of the temperature words involved in them and on the formal relatedness between them (W 1 --W 2 -the words W 1 and W 2 are formally related by an explicit marker, W 1 -W 2 -the words W 1 and W 2 are related by conversion, W 1 W 2 -the words W 1 and W 2 are not formally related).
Finally, there are questions on the constructions used for the different frames of temperature evaluation and the relations among them: are the constructions completely different, identical or in one or another way related to each other? Again, AMB_QR predications and TACT predications normally share the same construction (the standard adjectival or stative verb construction), so what primarily varies is the relation between AMB_NR predications and PF predications.
In some languages these two types of temperature predication may resort to completely different constructions. Palula presents such a case: as explained in connection with ex. (3) in Section 1, AMB_NR predication is basically an existential construction with or without an overt copula or with phrasal verbs like 'become', while PF predication involves an intransitive Expertum-oriented construction with verbs like 'falling' and 'rising' and experiencer marked with the Dative case.
In addition, there are also languages in which AMB_NR and PF predications are clearly related. In Bamana, AMB_NR predications also have the form of an existential construction (ex. (9a)). However, when the same construction is further expanded by a prepositional phrase with the locative/possessive marker, it is used for expressing both possession and PF predication (ex. (9b)). Ewe shows another version of a very close similarity between AMB_NR and PF constructions in that both involve the verb 'do/happen', used intransitively for AMB_NR (ex. (10a)) and transitively for PF temperatures (ex. (10b)). In terms of constructional approaches to language, further elaborated in Koch's (2012) "constructional typology", the two constructions in each of the examples (9) and (10) are related via inheritance links. Inheritance links are the constructional correspondence to the notion of formal relatedness, operating on the lexical level. In this terminology, the AMB_NR constructions in both Bamana and Ewe are related to the corresponding PF constructions via a subpart link, which is one of the four types of inheritance links in Goldberg's (1995) model. Figure 8 captures the information on the relations among the constructions involved in temperature predications in Palula by adding a third, constructional, layer to the morphosyntactic layer and the formal-relatedness layer and spelling out the details both for the lexical items and for the constructions. Figure 9 gives a more schematic and succint representation of the situation in Bamana and Ewe by using the abbreviations W n and Cx n instead of concrete lexical items and concrete constructions (and omitting the specification of the layers themselves). The arrow from Cx 3 to Cx 2 shows that Cx 2 is related to Cx 3 via an inheritance link.
In the languages listed in Table 2, the situation captured by Figure 9 is relatively infrequent. In other words, it seems that in languages that use the same temperature noun in both AMB_NR and PF predications tend to use them in very different constructions (i. e., not related to each other), but the basis for this observation is currently limited.
In Section 5, I will apply the same method to compare and represent temperature predications in some of the languages in which PF follow the Non-oriented intransitive strategy. In languages of this kind, PF and TACT predications involve  temperature terms that do not (necessarily) differ in their word-class affiliation, but may still differ in their morphosyntactic properties, which makes them different from the cases considered in this section.  Pustet 2015). As explained in Section 3 (cf. Table 1 and the text introducing it), Non-oriented PF constructions lack subjects or have an ambient (alt. dummy) subject. Their intransitive version involves an intransitive temperature denoting predicate, either verbal or non-verbal, and the Experiencer encoded as the non-core argument, typically in the Dative case (or with the marker having similar functions). Again, even though sharing the defining properties, the relevant constructions and the properties of the words involved in them are not the same across languages. In order to understand and represent the cross-linguistic variation in this case I will follow here the same procedure as in the preceding section and ask questions pertaining to three different levels of linguistic analysis -morphosyntax, formal relatedness, and constructions.

Languages in which personal-feeling temperature predications follow the Non-oriented intransitive strategy
Examples (11)-(13) show slightly different versions of the contrast between the standard adjectival predication for TACT and AMB_QR temperatures, and the Non-oriented intransitive strategy for PF temperatures, with the focus on the morphosyntactic properties of the temperature terms involved in them.
First of all, there are languages like Komi-Permyak (komi1269) and Komi-Zyrian (komi1268), both Uralic, in which the temperature term for 'cold' in both TACT and PF uses appears in exactly the same form, basically because adjectives lack agreement (ex. (11)). In other words, there is no morphosyntactic distinction between the temperature terms used in these two frames of temperature evaluation.
(11) Komi-Permyak (Ladygina 2012) a. Tactile: Ju-as va-ïs ködzyt river-loc.poss3sg water-poss.3sg cold 'The water in the river is cold.' b. Personal-feeling: Mujkö menïm ködzyt somehow I.dat cold 'I am somehow a bit cold.' The Komi case is contrasted to the Polish situation, illustrated in ex. (12). In Polish, as in all Slavic languages, PF temperatures are expressed by an impersonal construction involving the default form of the copula (3 rd person singular for nonpast and neuter singular for past) and the form called "adverb" or "predicative" in the Slavic linguistic tradition (Ru kategorija sostojanija). The temperature terms in two uses are therefore traditionally conceived of as being derivationally related, but belonging to two different word classes, which means that there is a clear morphosyntactic split between the temperature terms involved in TACT and AMB_QR predications, on the one hand, and those involved in PF predica-tions, on the other. However, as will be discussed later in this section, the decision to treat predicatives as a separate word class is not straightforward.
(12) Polish (Marcin Włodarczak p. c.) a. Tactile: Wod-a jest ciepł-a. water-nom.sg is warm-f.sg.nom 'The water is warm.' b. Personal-feeling: Był-o mi ciepł-o. was-n.sg me.dat warm-pred. 'I felt warm.' Finally, in Icelandic, the TACT (and AMB_QR) and the PF involve the same temperature adjectives, but with different morphosyntactic properties. Adjectives in the standard predicative construction appear in the nominative case and agree with the subject in gender and number, whereas PF temperatures are expressed by impersonal constructions with the 3 rd person singular of the copula and the default (neutral singular) form of the temperature adjective (ex. (13)). In other words, the temperature terms in the two uses have different inflectional potential, with the temperature adjective in PF constructions being restricted to just one form. I.dat is cold.n.sg.nom 'I feel cold.' The pattern illustrated in ex. (11), whereby Non-oriented intransitive PF predications and TACT predications involve the same uninflected adjective, is relatively rare in the languages represented in Table 3. In most cases there is a clear morphosyntactic difference between the two forms (PF temperatures are expressed by a default form of the adjective, as opposed to the agreeing forms used for TACT, as in ex. (13), or by a form, derivationally related to it, as in ex. (12)). Moreover, the form used for PF temperatures is also used for AMB_NR predications, confirming the pattern discussed in Section 4. 11 This is illustrated in (14) (cf. ex. (10b), (11b), and (12b)).
11 Both Latvian and Finnish allow two different forms for the temperature term in AMB_NR predications, of which only one occurs in PF predications. This is, however, in line with the pattern mentioned here, since the form used in PF predications is also allowed in AMB_NR predications.  However, on the whole, the morphosyntactic variation within temperature predication in languages with the intransitive Non-oriented PF strategy is more varied than in the cases considered in the preceding section, which always have a radical word-class split between the temperature terms involved in TACT and AMB_QR predications (temperature adjectives/stative verbs), and those involved in AMB_NR and PF predications (temperature nouns). Figure 10 compares the morphosyntactically based configurations attested in the languages in the two groups.
Let's now turn to the constructions used for the different frames of temperature evaluation and the relations among them. In the majority of the languages in Table 3, the constructions used for PF and AMB_NR predications are clearly related, differing only in the presence of the experiencer marked with the Dative case or with its language-specific equivalent. In other words, the AMB_NR constructions in these languages are related to the corresponding PF construc- tions via a subpart (inheritance) link. The only exceptions here are German and Icelandic, which have the ambient/dummy subject es/þat in AMB_NR predications (Bolinger 1973;Malchukov and Ogawa 2011: 25), as opposed to the subjectless impersonal PF predications, cf. ex. (13b) and (14b). In these two languages, the AMB_NR predications can be analyzed as being a particular instance of the standard adjectival predication (cf. ex. (13a)), used for both TACT and AMB_QR evaluation, i. e., being linked to it via an instance (inheritance) link in the Goldberg/Koch model. In other words, the situation here is in a way opposite to the one considered in Section 4, where constructions used for TACT/AMB_QR, AMB_NR and PF predications in most of the languages were not related to each other. Figure 11 compares the two constructionally based configurations attested for the languages of the two groups (i. e., those considered in this section and in Section 4).
Finally, when it comes to the temperature words themselves and to their formal relatedness, or motivational patterns, the patterns found here also parallel those considered in Section 4. In other words, as shown in Table 3, the temperature words in PF and AMB_NR vs. TACT predications stretch between being practically the same via clearly related but different to being completely unrelated. However, when the forms used in PF and AMB_NR vs. TACT predications are related to each other, these relations are considered to be inflection in some cases, but derivation in others, partly depending on the language-specific tradition and often without any further argumentation. In what follows I will devote some space to the role of predicatives in the Slavic temperature systems and elaborate on how three-layered semantic maps may be useful for capturing cross-linguistic variation in these systems.
The status of Slavic predicatives as inflectional vs. derivational forms and, consequently, their word-class status is disputed, since they are far from both canonical inflection and canonical derivation (cf. Corbett 2010;Spencer 2014: 58-63). In different sources they are therefore classified as either adverbs or as (short) neuter singular forms of adjectives or as a separate word class, partly depending on the language-specific tradition (cf. Letuchij 2017 andCorbett 2004: 207 on Russian and Zimmerling 2018 on Slavic). 12 Importantly, in most cases there is a completely straightforward relation between the temperature adjective and the corresponding predicative, as illustrated by the detailed three-layered semantic map for Polish in Figure 12.
However, the relation between the temperature adjective and the predicative is sometimes less straightforward; the three particularly interesting cases will be considered below (cf. Koptjevskaja-Tamm 2022 on the Slavic temperature systems).
The first case concerns the adjective horký 'hot' in Czech, that has two corresponding predicatives -a regular predicative (horko) and another one, based on a completely different stem (vedro), cf. ex. (15). Since both predicatives show the same ending in -o and share the same morphosyntactic behaviour, vedro could be considered as the optional suppletive predicative version of horký within the regular (inflectional or derivational) paradigm. Figure 13: The detailed three-layered semantic map for 'hot', 'warm', and 'cool, cold' temperature predications in Czech (excluding "studený"/"zima" for 'cold').  Figure 13 represents the temperature system in Czech involving the three of the four central temperature adjectives (horký 'hot', teplý 'warm' and chladný 'cool, cold') and their predicative correspondences. The fourth temperature adjective, studený 'cold' will be treated separately immediately below. 13 The second case concerns the predicative 'cold' in Czech, Slovak and Upper and Lower Sorbian, zima, which is unusual in lacking the ending -o. It has the same form as the noun 'winter' and is, strikingly, related to the 'cold' adjective 13 Both Czech and Slovak have two 'cooling' adjectives -studený and chladný. The exact semantic difference between those is unclear, but in both languages chladný seems to be restricted to some registers and styles (Viktor Elšik for Czech and Pavol Štekauer for Slovak p. c.). Here chladný is glossed as 'cool, cold' as opposed to studený, glossed as 'cold', since the 'nice cool' nuance is regularly reported for the former, but not for the latter.
in Sorbian (zymny), but not to those in Czech or Slovak (studený). Moreover, at least in Czech, zima can still function both as a noun 'cold' and as a predicative in AMB_NR and PF predications. Ex. (16) shows an intransitive Expertum-oriented PF predication in which zima allows adjectival modification and requires gender agreement on the copula. In other words, here it behaves as a proper temperature noun and the whole predication follows the strategy considered in Section 4 (in this case an existential construction expanded by the experiencer in the Dative case). Ex. (16b) is an impersonal construction with the predicative, that (marginally) allows adverbial modification and has a default (neuter) gender form of the copula.  Viktor Elšik (p. c.) mentions several other colloquial expressions for AMB_NR and PF temperatures involving nouns and the corresponding noun-based predicatives: the somewhat outdated hic from German Hitze 'heat' (ex. (17a)-(17b)) and several expressions for 'cold' (ex. (17c)-(17d)). According to him, the use of impersonal construction (in which the copula appears in the default form) is more common for the PF predications, conforming to the provisional generalization suggested in Section 4 that the use of a temperature noun in PF predications implies its use in AMB_NR predications. The comparison between the two alternatives in ex. (16) and (17) suggests a tentative historical process by which an erstwhile noun gradually changes its morphosyntactic properties to fit the general construction for a particular function. Evidently, without considering historical data these suggestions remain highly speculative. Figure 14 shows the semantic map for the temperature predications based on studený/ zima in Czech and the direction of suggested constructional reanalysis (by thick arrows). In this map the predications with the temperature adjectives are placed above the others to make the representation more reader friendly. Figure 14: The three-layered semantic map for the 'cold' ("studený" -"zima") temperature predications in Czech (thick arrows show the direction of constructional reanalysis).
In the two Czech cases considered above the distinction between temperature terms based on different lexical stems is intimately related to their morphosyntactic properties: the split occurs therefore between the temperature adjectives used for TACT and AMB_QR predications, and the temperature predicatives/nouns used for AMB_NR and PF predications. The third Slavic case to be considered here is Russian, in which the lexical distinction between the tactile (gorjačij/ gorjačo) and the non-tactile (žarkij/ žarko) 'hot' terms cuts across the morphosyntactic distinction and is therefore more consistent from the semantic point of view (cf. Koptjevskaja-Tamm 2022 on the development of this distinction in Russian). In ex. (18), the adjective gorjačij is used exclusively for TACT, while AMB_QR temperatures are described by the adjective žarkij, related to the predicative žarko, used for AMB_NR and PF temperatures. Moreover, the tactile adjective also has a predicative version, as seen in ex. (18), involving what is called here a "tactile experiencer". What is evaluated here is not the experiencer's "internal state" (as the case is with the PF predications), but something external, the temperature of which can be estimated in a tactile way. This sentence would be appropriate if the water with which Petja is washing his hands is too hot, or when he has touched a very hot surface.
(18) Russian: gorjačij/gorjačo vs. žarkij/žarko 'hot' in predication (own data) a. Tactile Figure 15 shows the semantic map for temperature predications in Russian (the tactile experiencer subframe excluded). Summarizing the presentation of the temperature systems in Polish, Czech and Russian, we can see how the three-layered semantic maps in Figures 11-14 can assist in visualizing the similarities and the differences among them. The Polish and the Czech temperature systems are basically identical on all the three levels for three of the four temperature values ('hot', 'warm', 'cool/cold'), with the additional suppletive predicative for 'hot' (Figures 12-13). For the 'cold' value in Czech there are two options: one of them (Figure 14b) builds on the predicative for the AMB_NR and PF predications and differs from the Polish system and the rest of the Czech system only on the level of lexical relatedness (the temperature predicative is not related to the temperature adjective). The other option (Figure 14a) has the same configuration as the first one, but differs from it in the details on the morphosyntactic and constructional layers. The Polish and the Russian systems ( Figure 12 and Figure 15) are identical on all the three levels for three of the four temperature values ('warm', 'cool', 'cold'), the only difference being the lexical split between the tactile and the non-tactile 'hot' for Russian. Crucially, the configurations shown in these maps are not conditional on the decision whether to treat predicatives as being an inflectional form of the corresponding adjectives or as being related to them via derivation.
The consistent lexical distinction between tactile and non-tactile uses (covering all AMB and PF uses), is not unique for Russian, but has cross-linguistic correspondences. Among the languages in Table 3 it is made for the 'hot' terms in Komi (pïm vs. žar, the latter borrowed from Russian) and for both 'hot' (kavrəm vs. ruvəŋ) and 'cold' (pɔtəm vs. iśk'i) in another Uralic language, Khanty (khan 1243), cf. ex. (19).
(19) Khanty: pɔtəm vs. iśk'i 'cold' in predication (Ladygina 2012 and Nadezhda Grishkina p. c.) a. Tactile: kăt'-əm n'uł-eł pɔtəm cat-1sg nose-3sg cold_tact 'The cat's nose is cold (to touch)' b. Quasi-ref.ambient: ipuš toveł vǒ-s šeŋk iśk'i once Spring be-pst.3sg very cold_nontact 'Once the Spring was very cold.' c. Non-ref.ambient: nǒmən (śik'em) iśk'i above (very) cold_nontact 'It is very cold in the attic.' d. Personal-feeling: man-em(a) iśk'i 1sg-dat cold_nontact 'I feel cold.' The final semantic map in this section ( Figure 16) shows the temperature system in Khanty. A comparison with the Russian system, shown in Figure 15, shows that the two languages clearly differ on the morphosyntactic level (Khanty lacks morphosyntactic differences between the temperature terms used in the different frames of evaluation), are identical on the constructional level, and share certain similarities on the level of lexical relatedness in having a lexical split between tactile and non-tactile temperature terms for some of the temperature values (only for 'hot' in Russian, but for both 'hot' and 'cold' in Khanty).

Conclusions
As noted in the beginning of this article, lexical typology has to a large extent been interested in lexical categorization of various cognitive domains and in lexical motivation, whereas grammatical behavior of words and morphosyntactic patterns as encoding meanings traditionally belong to grammatical typology. In this study I have attempted to show that the cross-linguistic variation in how languages carve up the domain of temperature cannot be accounted for without considering the interplay between the choice of a particular lexeme and its use in particular constructions. The two leading questions for the study, repeated from Section 1, have been how the lexicon and grammar interact with each other in the expression of the distinctions among the different frames of temperature evaluation across languages, and how this interaction can be represented. I have inspected a number of examples from different languages and have made a case for why the relations among the temperature words involved in predications about tactile, quasi-referential ambient, non-referential and personal-feeling temperatures need to be analyzed from three different perspectives: -their formal relatedness (i. e., whether they are identical, completely unrelated, related via derivation or inflection), -their morphosyntactic properties (e. g., their part-of-speech affiliation, inflectional potential, etc.), and -the constructions they occur in.
I have argued that these relations can be captured by an elaborated version of classical semantic maps, with three layers corresponding to the three perspectives above. In general, the cases considered in the article confirm the basic configuration of the frames of temperature evaluation, as proposed in my earlier work (Koptjevskaja-Tamm 2011;2015: 17): TACT -AMB_QR -AMB-NR -PF. I have showcased a number of examples to illustrate some of the ways in which the three layers contribute to the categorization of the domain and demonstrated that the contiguity is maintained on each of them. As argued in Koptjevskaja-Tamm (2015: 17), the motivation for this lies in the conceptual and perceptual affinities of ambient temperature with both tactile and personal-feeling temperatures. On the one hand, tactile and ambient temperatures are experience-based -they are about temperatures that can be verified from "outside". This makes them different from personal-feeling temperatures, which are experiencer-based, i. e., are about a subjective "inner" experience of a living being. But on the other hand, ambient and personal-feeling temperatures are rooted in the same type of experience, thermal comfort, whereas tactile temperature relates to evaluation of the temperature of other entities based on perception received by the skin. And quasi-referential ambient temperatures are closer to tactile temperatures than non-referential ambient temperatures in pertaining to particular entities.
The enterprise of creating three-layers semantic maps requires a number of analytical decisions that may be coupled with various methodological challenges.
First of all, as in all classical semantic maps, it is necessary to identify the nodes (meanings, functions) relevant for the organization of the semantic domain. All the maps considered in this paper case of temperature are based on four frames of temperature evaluation (tactile, quasi-referential ambient, non-referential ambient, personal-feeling temperatures), whereas most of the earlier research on temperature, including my own (e. g., Lehrer 1970;Prator 1963;Goddard and Wierzbicka 2007;Koptjevskaja-Tamm 2011;FrameNet) basically lumps together the two kinds of ambient temperatures (cf. Section 1). The decision to distinguish the two was formed in response to the growing crosslinguistic evidence justifying it.
Second, it is not apriori obvious which morphosyntactic categories and properties are relevant for the categorization of the semantic domain. Obviously, exemplary word classes distinctions are easy to spot -a noun and a verb are normally two different words. But in many of the cases considered in the paper the morphosyntactic distinctions are much more subtle, as, for instance, the difference between a fully inflectional adjective and a word that may either be considered as one of its forms or as a word derived from it. In this study I have chosen representations that are sensitive to a broad spectrum of morpho-syntactic distinctionsfrom the major word-class distinctions of the former type to the fairly fine-grained distinctions of the latter.
Third, the issue of formal relatedness is not always apparent either. On the one extreme, there are challenges with historical relatedness (the extreme case of loose colexification). Do we want to indicate that the two 'hot' adjectives in Russian, gorjačij (tactile) and žarkij (non-tactile) go back to the same Indo-European stem * g u̯ her-, but with different vowels? Probably not. But considering more transparent cases, does it matter -and for whom and for what -that the 'cold' adjective (šidáalu) and the 'cold' noun (šid) in Palula are historically related, as contrasted to the 'hot' adjective (táatu/téeti) and the 'hot' noun (húluk)? On the other extreme, the distinction between strict and loose colexification, or the one between one and the same lexeme and different lexemes, is also often tricky. 14 Are the different manifestations of Italian freddo 'cold' discussed in Section 4 and shown in Figure 7 one and the same lexeme with both adjectival and nouny properties, or are there two different lexemes related by conversion? Are the Slavic predicatives and the corresponding adjectives one and the same lexeme or are they related by derivation?
Finally, identifying constructions and their relations is, again, a delicate issue. To come back to the first examples considered in this paper: how should the constructions used for AMB_NR predications in English (ex. (1c)) and Wolof (ex. (2c)) be analyzed? They are basically the same as those used for all the other temperature predications, except for the invariant ambient/dummy subject, which probably justifies their treatment as connected to the others via an inheritance link of the 'instance' type. I leave the details of graphic implementation of these relations on a semantic map without violating the connectivity requirement for the future. Now, it is unsurprising that semantic maps of the kind suggested here are beneficial for capturing the temperature system of a particular language, but what about cross-linguistic comparison? It is true that fully elaborated three-layered semantic maps like those in Figure 8 and Figures 12-16 are simply too detailed and may be -and most probably often are -unmanageable in cross-linguistic comparison. This is also a problem inherent in most constructional approaches to language description and analysis, which often result in meticulous languagespecific networks, hardly generalizable to other languages. However, as illustrated throughout Section 5, cross-linguistic comparison can advantageously be done on the basis of more schematic representations, like those in Figures 9-11, which focus on particular configurations on particular levels (e. g., particular morphosyntactic or constructional configurations) and leave out the more specific details. The option of using more detailed or more schematic representations for specific purposes is of course not completely novel, but is very much in the spirit of what typologists always do, given the general tension between the richness of their language-specific data and the leading concern to reduce them to manageable patterns. A parallel near at hand within the area of semantic maps and constructions is found in Traugott's (2016) study of modals, where she suggests to combine schema-construction maps to represent relationships between abstract, conceptual schemas linked to form, with micro-construction maps to represent relationships between specific constructions in the expression of.
On a closing note: I hope to have shown the advantages of applying threelayered semantic maps to study how different means jointly carve up the temperature domain, the task in which the lexical, morphosyntactic and constructional distinctions are often inseparable from each other. It is reasonable to ask if the issue of joint linguistic expression of categories is particularly relevant for the temperature domain (and also for the super-domain of location-existence-possession considered in Koch (2012) (cf. Section 3), to the exclusion of others. In other words, whether the practical application of three-layered semantic maps is basically restricted to one single domain. At the present I lack any clear answer to this question -after all, most of the lexical-typological research has been successfully done on the level of lexemes, without messing up with their grammatical or constructional behaviour. On the other hand, there is also research which aspires to reconcile the lexical and grammatical interests. For instance, one of the leading issues in the project on categorization of the cut and break domain (Majid and Bowerman 2007) has been the interface between syntax and lexical semantics, i. e., to what extent and how the argument structure properties of a verb are predictable from its meaning. For the aqua-motion domain, Koptjevskaja-Tamm et al. (2010) argue that the different lexicalizations in the Swedish-Dutch-German systems correspond to the constructional differences in the lexically much much less elaborated Polish-Russian systems (cf. Rakhilina et al. 2022 for the basic information on the aqua-motion domain). It is my contention that research in both grammatical and lexical typology will benefit greatly from a more careful examination of lexicon-grammar interaction across languages, and that multiple-layered semantic maps may be a suitable tool in this endeavour (cf. Koptjevskaja-Tamm and Veselinova 2020 for research questions common to lexical typology and morphology).   Notes: *Algerian Arabic has additional constructions for PF temperatures, not considered here. **Romanian uses constructions with the noun frig as the alternative for the adjective based amb-nr and pf predications for 'cold' (cf. Table 3 for the other more regular alternatives).   Notes: *Romanian uses constructions with the noun frig as the alternative for the adjective based amb-nr and pf predications for 'cold' (cf. Table 2). **zima 'cold' in amb-nr and pf predications in Czech can function both as a noun and as a predicative (cf. ex. (16) and Figure 11 in Section 5). Pustet (2015) Wolof Atlantic, Niger-Congo Senegal nuc1347 Jean Léopold Diouf p. c., Perrin (2015), Robert (2010), Creissels et al. (2015)