The noncausal/causal alternation and the limits of ambitransitivity in a sample of sub-Saharan languages

The first part of the present article concentrates on noncausal/causal pairs whose noncausal mem - ber is a monovalent verb referring to a process typically undergone by concrete inanimate entities. The cross-linguistic variation in the coding of such pairs is analyzed in a convenience sample of 30 sub-Saharan languages from various genetic units. 10 of the 30 languages show an extremely high degree of preference for the ambitransitivity strategy in the coding of this particular type of noncausal/causal pairs, whereas none of them shows such a high degree of preference for any of the other possible strategies. The second part of the article concentrates on the 10 languages of the sample making a systematic use of the ambitransitivity strategy for the type of noncausal/causal pairs inves - tigated in the first part of the article. It analyzes the extension of the ambitransitivity strategy to the coding of other semantic types of noncausal/causal pairs and to the coding of the psych alternation. In the questionnaire used to test the extension of the ambitransitivity strategy, a sizeable proportion of ambitransitive pairs is found in only three languages. In the others, the lack of any systematicity is the situation that predominates for the semantic types of noncausal/causal pairs other than that inves - tigated in the first part of the article and for pairs of psych verbs.


Introduction
In this article, in conformity with the terminology defined in the introduction to this special issue, the labels "noncausal" and "causal" refer to purely semantic notions, as opposed to "causative" (which implies derivation from noncausal to causal) and "decausative" (or "anticausative", which implies derivation from causal to noncausal)."Noncausal" and "causal" are relative notions: a verb (or predicate) is not noncausal or causal in the absolute but only in relation to another verb with which it forms a noncausal/causal pair (Haspelmath 2016).For example, show is causal in relation to see but noncausal in relation to make show.
Five types of strategies may be involved in the coding of noncausal/causal pairs of verb meanings: 1 ambitransitivity strategy: in an ambitransitive pair, there is no formal difference between the noncausal verb and its causal counterpart; decausativization strategy: in a decausative pair, the noncausal verb can be analyzed as formally more complex than its causal counterpart; causativization strategy: in a causative pair, the causal verb can be analyzed as formally more complex than its noncausal counterpart; equipollence strategy: the two members of an equipollent pair are formally related, but none of the two can be characterized as morphologically more complex than the other; this definition embraces several subtypes that Nichols et al. (2004) designate as double derivation, conjugation class change, auxiliary change and ablaut; suppletion strategy: in a suppletive pair, the formal difference between the noncausal verb and its causal counterpart cannot be analyzed as a particular instance of some more or less regular pattern.
Decausativization and causativization share a morphological orientation of the relationship between the two members of the pair and, consequently, can be grouped together as directed strategies, contrasting with the other three strategies, which can be characterized as undirected.Some comments are in order here about the equipollence strategy, since the distinction between ambitransitivity and equipollence may be problematic.The point is that not all the changes affecting inflection or auxiliaries that may be observed between the two members of a noncausal/causal pair are worth being considered as instances of the equipollence strategy.When they constitute lexical properties of verbs, the analysis of such changes as involved in the coding of the noncausal/causal alternation is uncontroversial.By contrast, in languages in which verbs do not inflect in the same way, or select distinct auxiliaries, in transitive and intransitive clauses, a change in inflection or auxiliaries automatically triggered by the intransitive or transitive nature of the clause constitutes a mere instance of transitivity marking, which by itself is compatible with the analysis of noncausal/causal pairs as ambitransitive pairs, if no other formal distinction can be observed between the clauses projected by the two members of the pair.Among the languages quoted in this article, this observation applies to Basque (where, whenever an auxiliary is required, it is automatically 'be' in intransitive clauses and 'have' in transitive clauses) and to three Mande languages: Mandinka, Bambara and Soninke (where the paradigm of TAM markers is not exactly the same in transitive and intransitive clauses).
Still concerning the equipollence strategy, it is reasonable to think that, in general, taking into account the distinction between the different subtypes identified by Nichols et al. (2004) would give a finer-grained and richer understanding of the strategies involved in the noncausal/causal alternation.
However, there is no reason to think that this could have an impact on the question of the extension of the ambitransitivity strategy, which constitutes the main focus of this article, and this is the reason why I will not pursue that in this article.
Concerning now the distinction between equipollence and suppletion, I adopt a relatively broad view of the notion of suppletion, according to which pairs showing partial formal resemblance, but for which no regularity can be found in the formal relationship between the two members of the pair, are classified as suppletive rather than equipollent.This applies, for example, to Soninke kàrá/kàrí 'die, go out/kill, put out (fire)', analyzed as a suppletive pair because the verbal lexicon of Soninke includes no other pair of semantically related verbs showing the same formal difference.
In the remainder of this article, the first three letters of each of the labels put forward above will be used as an abbreviation for the five strategies that may be involved in the coding of the noncausal/ causal alternation: amb for ambitransitivity, dec for decausativization, cau for causativization, equ for equipollence and sup for suppletion.As illustrated in (1), for some noncausal/causal pairs of verb meanings at least, the five possible strategies are attested cross-linguistically, which raises the question of possible regularities in the choice of particular strategies by individual languages.
(1) 'go out/put out (fire)' in five sub-Saharan languages a. Minyanka (Gur) The regularities in the choice of particular strategies by individual languages and in the preference of individual noncausal/causal pairs of verb meanings for the causativization or decausativization strategy have already been discussed in very general terms by Haspelmath (1993), Nichols et al. (2004) and Haspelmath (2016).One of the hypotheses suggested by the data they analyze is that, in a given language, different semantic types of noncausal/causal pairs do not necessarily show the same possibilities or preferences.This question is addressed here on the basis of a convenience sample of 30 sub-Saharan languages from various genetic units, focusing on the use of the ambitransitivity strategy.Given the importance of the ambitransitivity strategy in many languages from sub-Saharan Africa, such a sample is well suited for discussing the extension of this particular strategy beyond the semantic type of noncausal/ causal pairs for which it is particularly common.
The first part of this article (Sections 2 to 5) deals specifically with pairs whose noncausal member is a monovalent verb referring to a process (not a state) typically undergone by concrete inanimate entities and easily conceived as occurring without the involvement of a clearly identified external instigator.This choice is motivated by the fact that, cross-linguistically, in comparison with other semantic types of noncausal/causal pairs, this type seems to show the following two particularities: on the one hand, a relatively high proportion of pairs whose members are either morphologically related or identical and, on the other hand, a particularly important cross-linguistic variation between the possible strategies (cf. in particular Haspelmath 2016).
Section 2 presents the questionnaire I used to evaluate the cross-linguistic variation in the formal relationships between noncausal verbs meeting the semantic characterization formulated above and their causal counterparts (Questionnaire A).Section 3 describes the language sample.Section 4 compares the coding of the noncausal/causal pairs from Questionnaire A across the languages of the sample.Section 5 discusses the prevailing tendencies in the coding of individual noncausal/causal pairs.The analysis carried out in Sections 4 and 5 shows that the language sample includes a high proportion of languages (10 out of 30) in which the verb pairs that constitute Questionnaire A show an extreme degree of preference for the ambitransitivity strategy but no language showing a very high degree of preference for either the causativization or decausativization strategy.Consequently, this sample does not provide a convenient basis to discuss the extension of the causativization or decausativization strategy to other semantic types of noncausal/causal pairs but can be used to discuss the following question: in the languages that strongly prefer the ambitransitivity strategy for the semantic type of noncausal/causal pairs investigated in the first part of the article, to what extent is it possible to predict the extension of this strategy to other semantic types of noncausal/causal pairs or to the psych alternation (a type of valency alternation often treated as a mere subtype of the noncausal/ causal alternation)?
In the second part of the article (Section 6), I use a second questionnaire (Questionnaire B) to evaluate the tendencies in the coding of semantic types of noncausal/causal pairs other than that investigated in the first part of the article and in the coding of pairs of psych verbs.The data collected on the basis of Questionnaire B shows that the strong preference for the ambitransitivity strategy observed in some languages with the noncausal/causal pairs that constitute Questionnaire A does not necessarily extend to other semantic types of noncausal/causal pairs or to the psych alternation (Section 6).
Section 7 summarizes the conclusions.

Questionnaire A
As explained in the introduction, among the possible semantic types of noncausal/causal pairs, the first part of the present article deals specifically with pairs whose noncausal member is a monovalent verb referring to a process (not a state) typically undergone by concrete inanimate entities and easily conceived as occurring without the involvement of a clearly identified external instigator meeting the definition of a prototypical agent.Taking into account this delimitation, the wordlists already used for similar studies (Haspelmath 1993;Nichols et al. 2004), and my own experience of working on sub-Saharan languages and of consulting dictionaries of sub-Saharan languages, I selected the 13 pairs of verb meanings listed in (2).In order to facilitate comparison with other studies of valency orientation, I decided to restrict the questionnaire to noncausal/causal pairs that also feature, or at least have a near equivalent, in the questionnaire used by Haspelmath (1993: 97), which served as a basis for many subsequent studies.
( As illustrated in (3), 2 the relevance of this questionnaire for a cross-linguistic investigation of noncausal/causal pairs follows from the fact that, in some languages, the noncausal/causal pairs that constitute it show an exclusive (or almost exclusive) preference for one of the possible strategies, for example ambitransitivity in Basque (unclassified, eus), with 13 ambitransitive pairs out of 13, decausativization in Romanian (Romance, rum), with 12 decausative pairs out of 13, 3 and causativization in Akhvakh (Nakh-Daghestanian, akv), with 11 causative pairs out of 13. 4   (3)

The language sample
The language sample includes 30 languages belonging to 15 distinct genetic units (either independent language families or subfamilies of one of the three phyla variously recognized by specialists of the historical study of sub-Saharan languages -Afroasiatic, Nilo-Saharan and Niger-Congo). 5 (4) The language sample and the sources Genetic Unit Glossonym (ISO 639-3 code) Source Atlantic Balant Ganja (bjt) own data Adamawa Fula (fub) Noye (1990) This is a convenience sample.In addition to the sub-Saharan data already published in the World Atlas of Transitivity Pairs (Herero, Swahili, Sidaama, Kupsabiny, Amharic), 6 my own data on Balant Ganja, Jóola Fóoñi, Tswana, Mandinka and Soninke, and data provided by experts of other languages, I consulted all the dictionaries of African languages to which I had relatively easy access, and the languages in the sample are simply those for which the following two conditions were met: (a) I was able to fill Questionnaire A without any gaps, and (b) in the case of languages with more or less complex morphophonological processes, I had the morphological information necessary to characterize the pairs as belonging to one of the five basic types without the risk of errors.

Introductory remarks
The coding of the 13 noncausal/causal pairs of Questionnaire A in the 30 languages of the sample is summarized in Appendix 2. Within the limits of the sample, the average value for each of the five possible types of strategies (i.e. the sum of the values for each of the languages divided by 30) is as indicated in (5).
6.Not all the sub-Saharan data provided by the World Atlas of Transitivity Pairs has been included in the sample.Data whose analysis I considered doubtful has not been included.In all cases, I have reevaluated the analysis, and the discussion of the languages in question in the present article refers to my own evaluation (see Appendix 1).
(5) Average values for each of the five possible types of strategies ambitransitivity 5.1 out of 13 decausativization 3.2 out of 13 causativization 2.6 out of 13 equipollence 1.5 out of 13 suppletion 0.5 out of 13 The suppletion strategy is found in a sizeable proportion of the languages of the sample for just one pair: go out/put out (fire) (9 languages out of 30).The average value for the suppletion strategy is particularly low, and none of the languages in the sample has more than 3 suppletive pairs.Consequently, the suppletion strategy will not be considered in the characterization of the individual languages in terms of relative prominence of particular strategies.Within the limits of the sample, I consider each of the other four strategies as relatively prominent in a given language if its value in the language in question exceeds the average value by more than one third, which gives the following thresholds: 7 -7 for the ambitransitivity strategy, -5 for the decausativization strategy, -4 for the causativization strategy, -3 for the equipollence strategy.

The relative prominence of the four main strategies in the languages of the sample
Most of the languages of the sample can be viewed as having one (and only one) relatively prominent strategy for the coding of the semantic type of noncausal/causal pairs represented in Questionnaire A.
In some others, two strategies show a relative prominence.In yet others, none of the strategies exceeds its average value significantly.This third group consists of the following four languages: Wolof (Atlantic), Soninke (Mande), Koroboro Senni (Songhay) and Humburi Senni (Songhay).

Languages with one relatively prominent strategy
10 languages belonging to 7 of the 15 genetic units represented in the sample show a relative prominence of ambitransitivity only: Emai (Benue-Congo), Sar (Central Sudanic), Jamsay (Dogon), Minyanka (Gur), Baule (Kwa), Fon (Kwa), Bambara (Mande), Kakabe (Mande), Mano (Mande) and Gbaya (Ubangian).It is remarkable that all of them have a very high proportion of ambitransitive pairs, of the same range as that found, for example, in English (between 10 and 12 out of 13).7 languages belonging to 5 of the 15 genetic units represented in the sample show a relative prominence of decausativization only: Jóola Fóoñi (Atlantic), Seereer (Atlantic), Lingala (Benue-Congo), Tswana (Benue-Congo), Sidaama (Cushitic), Kupsabiny (Eastern Sudanic) and Kanuri (Saharan).The number of decausative pairs is particularly high in Kupsabiny: 10 out of the 13 pairs of Questionnaire A, i.e. exactly the same number as in Russian.Other languages with a marked prominence of the decausativization strategy include Jóola Fóoñi (8 decausative pairs), Seereer (9 decausative pairs) and Kanuri (8 decausative pairs).This is all the more remarkable because it has been repeatedly stated in the literature on the noncausal/causal alternation (cf.among others Haspelmath 1993: 102;Nichols et al. 2004: 182) that a strong preference for decausativization is extremely rare (or even inexistent) outside of Europe.
The sample includes just one language characterized by the relative prominence of causativization only: Swahili (Benue-Congo).Note that, even in Swahili, the number of causative pairs in Questionnaire A is not very high: 7 out of 13, to be compared with the 8.5 causative pairs found in 7. I am conscious of the arbitrariness of considering a strategy as prominent in a given language if its value exceeds the average value by more than one third rather than any other percentage, but the data would be difficult to analyze consistently without first taking such a decision.In any case, a different decision on this point would have no significant impact on the discussion of possible regularities in Section 4.3.
Mongolian (Haspelmath 1993: 117-118) or the 11 causatives pairs found in Akhvakh (see Section 2) on the basis of the same questionnaire.
The sample also includes five languages with a proportion of causative pairs of the same range as that found in Swahili (between 5 and 7) but in which another strategy can also be viewed as relatively prominent: Amharic (Semitic), Sandawe (unclassified), Mandinka (Mande), Herero (Benue-Congo) and Afar (Cushitic).
Two languages show a relative prominence of the equipollence strategy only: Adamawa Fula (Atlantic) and Hausa (Chadic).Unsurprisingly, a salient characteristic shared by Fula and Hausa is the existence of a system of conjugation classes (or inflectional voices).Hausa has several verb classes (traditionally called "grades") that differ primarily in the forms that verbs take depending on their objects or lack of objects. 8Fula has three verb classes (traditionally called "voices") that differ in the forms of the TAM and polarity markers.In both cases, conjugation class change is a common way of coding the noncausal/causal alternation. 9A system of conjugation classes (or inflectional voices) is also found in Balant Ganja (Atlantic), but in Balant Ganja, the change in conjugation class observed in the coding of the noncausal/causal pairs is often the mere consequence of the addition of a decausativization marker selecting a particular voice, and the pairs in question are consequently best viewed as decausative pairs.
Among the languages that I did not include in the sample because of some gaps in the data, the predominance of the equipollence strategy is nevertheless obvious in Pere (Adamawa).However, the equipollent pairs of Pere do not involve conjugation class change but tone ablaut, as, for example, nál 'melt (intr.)'/nàl'melt (tr.) (Kastenholz 2014).

Languages with two relatively prominent strategies
Within the limits of Questionnaire A, Mandinka (Mande) has a relatively high proportion of both ambitransitive and causative pairs.Amharic (Semitic) and Sandawe (unclassified) have a relatively high proportion of both decausative and causative pairs (and consequently very few undirected pairs, or none at all).Balant Ganja (Atlantic) shows a relative prominence of both decausativization and equipollence.
None of the languages of the sample shows a relative prominence of both ambitransitivity and equipollence, in line with one of the correlations observed by Nichols et al. (2004: 165) in their worldwide sample.

The ambitransitivity strategy in the languages of the sample
As mentioned above, 10 out of the 30 sample languages have a very high proportion of ambitransitive pairs in Questionnaire A, and the predominance of ambitransitivity is also obvious for several languages that I did not include in the sample because of some gaps in the data: Samba Leko (Adamawa), Yoruba (Benue-Congo), Ikposo (Kwa), Bisa (Mande), Dan (Mande) and Sango (Ubangian).
In this respect, the sample of 30 sub-Saharan languages analyzed here sharply contrasts with the sample of 21 mainly Eurasian languages analyzed by Haspelmath (1993), among which English is the only language with such a high proportion of ambitransitive pairs.8.The reference grammars of Hausa consider some "grades" as basic and others as derived, but this distinction is based on semantic rather than formal considerations.As a rule, as discussed among others by Newman (2000), verb stems found in "derived" grades show a relatively constant element of meaning associated with the grade form which is absent in the "basic" grades.9. Note, however, that the vehicular/urban variety of Adamawa Fula tends to lose the middle voice and replaces it by the active voice, which results in a reduction of the proportion of equipollent pairs and an increase in the proportion of ambitransitive pairs (Henry Tourneux p.c.; Jean-Pierre Boutché p.c.).

The decausativization strategy in the languages of the sample
As mentioned above, at least for the semantic type of noncausal/causal pairs represented in Questionnaire A, some of the languages in the sample show a strong preference for decausativization.Among the languages for which I have partial data, the preference for decausativization is also obvious in Ibibio (Benue-Congo).
However, none of the languages in the sample shows a preference for decausativization comparable to that found in extremely decausativizing languages such as Romanian (3), and languages with no decausative pair at all within the limits of Questionnaire A are very common too (13 out of 30), which sharply contrasts with the scarcity of such languages in the language sample analyzed by Haspelmath (1993), heavily oriented towards Eurasian languages.
Consistent with Nichols et al.'s (2004: 166-167) observation that decausativization is favored by high morphological complexity, all the languages in the sample that have a relatively high proportion of decausative pairs are morphologically complex, whereas most of the languages that have a very low proportion of decausative pairs or no decausative pair at all within the limits of Questionnaire A are morphologically simple.Interestingly, the three languages in the sample that combine high morphological complexity and strong dispreference for decausativization (Fula, Herero and Hausa) also share a relatively high proportion of equipollent pairs.Finally, within the limits of the sample, the total lack of decausative pairs is particularly common among the languages showing a marked preference for ambitransitivity, which is consistent with the fact that, contrary to decausativization, ambitransitivity is favored by morphological simplicity.

The causativization strategy in the languages of the sample
As mentioned above, none of the languages in the sample shows a proportion of causative pairs in Questionnaire A comparable to that found in an extremely causativizing language such as Akhvakh (Nakh-Daghestanian) (3).The languages for which I have partial data confirm this observation.In some of them, for example Dida (Kru) and Bete (Kru), the prominence of the causativization strategy is obvious, but in all cases, the available data excludes the possibility of an extremely high proportion of causative pairs.Interestingly, the total lack of causative pairs within the limits of Questionnaire A can be found even in languages that have an otherwise relatively productive mechanism of morphological causativization.Bambara is a case in point.This means that, in the languages in question, causative derivation may be productive with other semantic types of noncausal/causal pairs but not with the type represented in Questionnaire A.
According to Nichols et al. (2004: 172), causativization prominence for inanimate noncausal/ causal pairs is found mainly in languages that do not have "passive or other A-removing or A-demoting processes".This generalization is clearly contradicted by the languages of the sample analyzed here.Among the languages of the sample that have a relatively high proportion of causative pairs, Mandinka has fully productive transitive-passive ambitransitivity, 10 and the other five (Herero, Swahili, Afar, Sandawe and Amharic) all have productive morphological mechanisms of passive and/or decausative derivation.
10.By fully productive transitive-passive ambitransitivity, I mean the possibility for any transitive verb to be also used in a zero-coded passive construction, as, for example, in Mandinka, tábì 'prepare (food)' in Fàatú mâŋ kínòo tábì 'Fatou did not prepare the meal'/Kínòo mâŋ tábì 'The meal was not prepared'.Nichols et al. (2004: 168-169) observe that, in their worldwide sample, ergative alignment is found in 6 out of the 9 languages they analyze as having high numbers of undirected inanimate verb pairs.On this basis, they put forward the following generalization: "The directed types significantly disfavor ergativity while undirected ones significantly favor it".

Valency orientation and alignment
None of the languages included in my sample of sub-Saharan languages shows ergative alignment, whereas in 12 out of 30, the proportion of undirected pairs exceeds 2/3.However, this cannot be viewed as contradicting Nichols et al.'s generalization about valency orientation and alignment, since they do not claim that undirected strategies are significantly disfavored in non-ergative languages.

Genetic relationships and the variation in the coding of noncausal/causal pairs
Three language families are represented in the sample by more than 2 members: Atlantic (5), Benue-Congo (5) and Mande (5).

The Benue-Congo family
Four out of the five Benue-Congo languages included in the sample (Herero, Lingala, Swahili and Tswana) belong to the Bantu group, a low-level genetic subgroup within the Benue-Congo family.Consequently, it is not surprising that, apart from the scarcity of decausative pairs in Herero, these four languages do not differ much in the coding of the noncausal/causal pairs included in Questionnaire A (with a high proportion of both causative and decausative pairs and very few undirected pairs), 11 whereas Emai, which belongs to another branch of Benue-Congo, shows a very different profile.Among the non-Bantu Benue-Congo languages for which I have partial data, the data is sufficient to characterize Yoruba as a language with a strong prevalence of ambitransitivity (like Emai), Degema as a language with a moderate prevalence of causativization and Ibibio as a language with a strong prevalence of decausativization.More research would need to be done before putting forward generalizations about valency orientation in Benue-Congo.

The Mande family
The five Mande languages included in the sample seem to be representative of the diversity across Mande languages.One of them (Soninke) has no particularly prominent strategy, another (Mandinka) shows approximately equal prominence of causativization and ambitransitivity, whereas the other three (Bambara, Kakabe and Mano) show a very strong prevalence of ambitransitivity.This last configuration seems to be particularly widespread among Mande languages.Among the Mande languages for which I have been able to find between 10 and 12 of the noncausal/causal pairs that constitute Questionnaire A, Bisa and Dan show a strong prevalence of ambitransitivity, whereas Soso seems to have a configuration similar to that of Mandinka, and Bozo and Bobo seem to have a configuration similar to that of Soninke.

The Atlantic family
The sample includes an Atlantic language with no particularly prominent strategy within the limits of Questionnaire A (Wolof), two languages with a marked prevalence of decausativization (Jóola Fóoñi and Seereer), a language with a marked prevalence of equipollence and no decausative pair at all (Fula) and a language with a relative prominence of both decausativization and equipollence (Balant Ganja).Balant Ganja is also the only Atlantic language in the sample in which the number of causative pairs departs from the average significantly, with just one causative pair in Questionnaire A.

Genetic proximity, contact and the coding of noncausal/causal pairs
The question briefly commented on in this section is specifically addressed by Allassonnière-Tang et al. (this issue) for the Atlantic, Mande and Mel families.
It is clear that the language sample on which the present article is based is not sufficient for an in-depth investigation of the possible relationships between the typology of the coding of noncausal/ causal pairs in sub-Saharan languages, their genetic affiliation and their contact history.Unsurprisingly, closely related languages (such as Lingala and Tswana, Humburi Senni and Koroboro Senni, or Bambara and Kakabe) often show similar configurations.However, the sample also includes two cases of very closely related languages with significantly different configurations, which suggests that, in the history of languages, contrary to what has been repeatedly stated in the literature on valency orientation, the coding of noncausal/causal pairs may be subject to relatively abrupt changes, whatever the possible reasons for such changes.
The first such case is that of Fula and Seereer, which constitute a subgroup within the Atlantic family.Within the limits of Questionnaire A, Fula and Seereer have an approximately equal number of causative pairs, but Seereer shows a strong prevalence of the decausativization strategy and has no equipollent pair at all, whereas Fula shows a strong prevalence of the equipollence strategy and has no decausative pair at all.This contrast is quite obviously related to the fact that Fula morphosyntax combines valency-changing derivations with a system of conjugation classes semantically similar to that of Ancient Greek, whereas Seereer has a rich and productive system of valency-changing derivations but no system of conjugation classes.Unfortunately, I am not aware of any evidence that could help to reconstruct the historical scenario responsible for this contrast.
The second case is that of Bambara and Mandinka, two Mande languages whose genetic closeness is so obvious that they are often presented as two dialects of the same macro-language.Within the limits of Questionnaire A, Bambara has 12 ambitransitive pairs and no causative pair, whereas Mandinka has 7.5 ambitransitive pairs and 5.5 causative pairs.Moreover, the relatively high number of causative pairs found in Mandinka, unusual for a Mande language, has no obvious explanation in terms of language contact, in spite of the fact that Mandinka differs from the other Mande languages by the importance of its contacts with Atlantic languages.The point is that all the Atlantic languages in the sample show a proportion of causative pairs in Questionnaire A lower than that found in Mandinka; in fact, contact with Atlantic languages should rather have favored the emergence of decausative pairs, which are not attested at all in Mandinka.
An even sharper contrast can be found between Emai and Degema, which both belong to the Edoid group within the Benue-Congo family.Degema is not included in the sample because of some gaps in the data, but the partial data I have for this language includes 4.5 decausative pairs, 5.5 causative pairs and no undirected pair, whereas Emai (included in the sample) has 12 ambitransitive pairs and 1 suppletive pair.

Prevailing tendencies in the coding of individual noncausal/causal pairs
For each of the pairs of verb meanings included in Questionnaire A, the number of languages of the sample selecting each of the five possible strategies is given in ( 6). ( 6

Variation in the ratio of undirected to directed pairs
For the vast majority of the pairs included in Questionnaire A, the ratio of undirected to directed pairs within the limits of the sample is comprised between 14/16 and 18/12, with an average value of approximately 16/14.Only two pairs show a ratio significantly different from the average: 'dry (intr./tr.)' and 'go out/put out (fire)'.They are examined in Sections 5.1.1 and 5.1.2,before examining the ratio of decausative to causative pairs in the eleven pairs showing an average ratio of undirected to directed pairs (Section 5.2).

The case of 'dry (intr./tr.)'
The ratio of undirected to directed pairs is much lower for 'dry (intr./tr.)'than for any of the other pairs included in Questionnaire A: 7.5/22.5.Another striking characteristic of 'dry (intr./tr.)' is an extremely low ratio of decausative to causative pairs (1/21.5).Interestingly, in the sample of 21 mainly Eurasian languages analyzed by Haspelmath (1993), 'dry (intr./tr.)' also shows a particularly low ratio of decausative to causative pairs (3/10), but its ratio of undirected to directed pairs (7/13) has nothing exceptional in comparison with that found for other pairs; cf. for example 5.5/15.5 for 'melt (intr./tr.)'.I have no explanation or hypothesis for this observation.

The case of 'go out/put out (fire)'
The ratio of undirected to directed pairs is much higher for 'go out/put out (fire)' than for any of the other pairs included in Questionnaire A: 23/7.This particularity of 'go out/put out (fire)' is not found in the sample analyzed by Haspelmath (1993) , in which the ratio of undirected to directed pairs for 'go out/put out (fire)' is 10.5/10.5, not very different from that of, for example, 'burn (intr./tr.)'(9/12).The very high proportion of undirected pairs for 'go out/put out (fire)' in the sub-Saharan sample is due to an unusually high number of languages that have a suppletive pair for 'go out/put out (fire)'.This is at least partly due to the fact that the colexification of 'go out/put out (fire)' with 'die/kill' is a common pattern among the languages of sub-Saharan Africa.The point is that the coding of 'die/ kill' by means of suppletive pairs is also particularly common in sub-Saharan languages, as in the languages spoken in other parts of the world, cf. for example the Koroboro Senni pair buu/wii 'die/ kill' and 'go out/put out (fire)' quoted in (1) or the Soninke pair kàrá/kàrí 'die/kill' and 'go out/put out (fire)'.Consequently, 'go out/put out (fire)' pairs that are originally a metaphorical extension of 'die/ kill' may have maintained the suppletive coding of the noncausal/causal alternation typically found for 'die/kill' pairs.

Variation in the ratio of decausative to causative pairs
The 11 pairs that show an average ratio of undirected to directed pairs can be ranked as shown in (8) according to the ratio of decausative to causative pairs. 12(9) gives the corresponding data for the Haspelmath sample.
(8) The ranking of the 11 verb meanings showing an average ratio of undirected to directed pairs according to the ratio of decausative to causative pairs dec/cau Ratio melt 1.5/13.5 0.11 burn 3/9 0.33 increase 6.5/9.5 0.68 spread 6.5/6.5 1

Variation in the preference for the ambitransitivity strategy
The 13 pairs of verb meanings that constitute Questionnaire A can be ranked as shown in (10) according to the number of languages from the sample that select the ambitransitivity strategy for each of them.Two groups of pairs of verb meanings of unequal importance can be distinguished: 11 pairs for which the ambitransitivity strategy is selected by a relatively high number of languages (between 11 and 14, i.e. more than half of the 21 languages in the sample that have at least one ambitransitive pair) and two pairs for which the ambitransitivity strategy is selected by a relatively low number of languages (about one third of the 21 languages in the sample in which the ambitransitivity strategy is attested).However, none of the pairs that constitute the sample shows an extreme degree of preference or dispreference for the ambitransitivity strategy.

Introductory remarks
In this section, I discuss the extension of the ambitransitivity strategy to semantic types of verb pairs other than that investigated in the previous sections.The languages selected for this study are the ten languages of the sample that strongly prefer the ambitransitivity strategy for the semantic type of noncausal/causal pairs considered in the first part of the article: Emai, Sar, Jamsay, Minyanka, Baule, Fon, Bambara, Kakabe, Mano and Gbaya.The investigation is carried out on the basis of the pairs of verb meanings listed in ( 11).
( The first type consists of noncausal/causal pairs whose noncausal member is a monovalent verb referring to a process typically involving non-volitional animate entities: die/kill, get tired/tire, get sick/make sick, heal/cure and wake up (intr./tr.).
The second type consists of noncausal/causal pairs whose noncausal member is a monovalent verb referring to a process typically involving volitional animate entities: get up/make get up, lie down/ make lie down, hide (intr./tr.), sit down/seat and stop (intr./tr.).
The third type consists of pairs of psych verbs referring to the same psychological event or state but differing in the perspectivization of the event/state, manifested in the mapping of the semantic roles of experiencer and stimulus onto syntactic roles: be afraid/frighten, be ashamed/put to shame, get angry/anger, get sad/sadden, hate/displease, like/please and rejoice/delight.
The first two types are not particularly problematic, but the third one calls for a brief comment, since pairs of psych verbs such as be afraid/frighten are often presented as noncausal/causal pairs in the literature.For example, Nichols et al. (2004) included the pairs fear, be afraid/frighten, scare and become angry/make angry in their questionnaire. 13My position on this point is that such pairs of psych verbs are not a subtype of noncausal/causal pairs for the simple reason that, whatever the possible similarities with the syntactic properties of noncausal/causal pairs, the semantic difference between the two members of psych verb pairs cannot be analyzed in terms of absence vs. presence of a participant occupying the initial position in the causality chain but only in terms of change in the perspectivization of the event.
The two members of psych verb pairs encode the same bivalent participant structure <Experiencer, Stimulus>.Moreover, a crucial contrast between psych verbs and the verbs involved in the noncausal/causal alternation is that none of the two participants in a psychological event can be 13.Interestingly, in contrast to the questionnaire used by Nichols et al. (2004), the questionnaire of 31 verb pairs used by Haspelmath (1993) for his cross-linguistic investigation of the noncausal/causal alternation includes no pair of psych verbs.
unambiguously characterized as having a particular affinity with one of two essential participants in a typical transitive event: typical agents are animate, like experiencers in psychological events, but as regards the involvement in the causality chain, it is the stimulus that has an affinity with agents in typical transitive events.
It is true that, in some languages, the equivalent of frighten derives from be afraid via the addition of a causative marker (as in Mandinka sílà/sílá-ndì).However, two uses of frighten should be distinguished.For example, The man frightened the child with his dog can be analyzed as the causative counterpart of The child was afraid of the dog, since an additional participant fulfilling the role of causer is involved.By contrast, the relationship between The dog frightened the child and The child was afraid of the dog cannot be analyzed in terms of causation, since The dog frightened the child and The child was afraid of the dog express two different perspectivizations of the same bivalent participant structure <Experiencer, Stimulus>: 'from Experiencer to Stimulus' in the case of The child was afraid of the dog and 'from Stimulus to Experiencer' in the case of The dog frightened the child.
In other words, in pairs such as Mandinka sílà 'be afraid'/sílá-ndì 'frighten', the causative suffix -ndi does not necessarily mark a change in participant structure but only in the perspectivization of a bivalent participant structure that remains unchanged.Interestingly, some languages are sensitive to this distinction.For example, in Basque, beldurtu corresponds to both be afraid and non-causative frighten, whereas the marked causative form beldurrarazi is preferred as the equivalent of causative frighten.
Before examining the sub-Saharan languages selected for this study, it is worth mentioning that, outside of Africa, the comparison of two languages such as English and Basque shows that the languages showing a very strong preference for ambitransitivity in the coding of noncausal/causal pairs of the semantic type investigated in the first part of this article do not necessarily behave in the same way with respect to the coding of the semantic types of verb pairs considered in this section.Within the limits of Questionnaire A, English and Basque show almost the same proportion of ambitransitive pairs (12 out of 13 for English, 13 out of 13 for Basque).By contrast, as can be seen in ( 12), Basque also has ambitransitive pairs for all of the pairs of verb meanings that constitute Questionnaire B, whereas within the limits of the same questionnaire, English has ambitransitive pairs for 5 pairs of verb meanings only. (

Extension and limits of ambitransitivity in the languages of the sample
In fact, the languages of the sample showing a very strong preference for ambitransitivity within the limits of Questionnaire A are very diverse in the treatment of the semantic types of verb pairs represented in Questionnaire B. The ambitransitive pairs I have found in Questionnaire B are listed in (13).

Languages with a strong prevalence of ambitransitivity in Questionnaire A and no ambitransitive pair at all in Questionnaire B
In two of the languages showing a strong prevalence of ambitransitivity in Questionnaire A, I was not able to find a single ambitransitive pair in Questionnaire B: Emai and Jamsay.
The source I used for Emai (Schaefer & Egbokhare 2007) is a very complete dictionary in which the possibility of using verbal lexemes transitively or intransitively is indicated with precision, and with the help of which I had no difficulty finding 12 ambitransitive pairs and 1 suppletive pair corresponding to the pairs of verb meanings that constitute Questionnaire A. By contrast, among the 17 pairs of verb meanings that constitute Questionnaire B, I was able to retrieve 9 pairs only, among which 3 suppletive pairs, 1 decausative pair, 1 equipollent pair and 4 causative pairs, but no ambitransitive pair at all.
In Jamsay, within the limits of Questionnaire B, I was able to retrieve 11 pairs out of 17: 10 causative pairs and 1 suppletive pair, but no ambitransitive pair at all.

Languages with a limited number of ambitransitive pairs in Questionnaire B
For 5 of the languages showing a strong prevalence of ambitransitivity within the limits of Questionnaire A, I was able to find a very limited number of ambitransitive pairs within the limits of Questionnaire B (one or two): Sar, Fon, Kakabe, Mano and Gbaya.In the languages in question, in addition to a limited number of ambitransitive pairs, I have found some suppletive pairs and also a limited number of pairs that can be analyzed as causative or decausative (although they involve periphrasis rather than systematic grammatical devices), but for most pairs, the dictionaries I have consulted (and with the help of which I had no difficulty in filling Questionnaire A) provide an equivalent for one of the members of the pair only.

Languages with a sizeable proportion of ambitransitive pairs in Questionnaire B
Among the languages of the sample showing a strong prevalence of ambitransitivity in Questionnaire A, three also have a sizeable proportion of ambitransitive pairs in Questionnaire B: Baule (10 out of 17), Bambara (7 out of 17) and Minyanka (6 out of 17).However, none of them shows an extension of the ambitransitivity strategy comparable to that observed, for example, in Basque (12).
In Baule, no other strategy emerges from Questionnaire B as relatively prominent, whereas Bambara and Minyanka also have a sizeable proportion of causative pairs (which were completely lacking in Questionnaire A).
Moreover, there is an interesting contrast between the two languages of the sample that show a strong predominance of ambitransitivity in Questionnaire A and a relatively wide use of both ambitransitivity and causativization in Questionnaire B. In Minyanka, ambitransitivity and causativization are not attested for the same pairs of verb meanings.By contrast, in Bambara, as a rule, ambitran-sitivity and causativization are attested as variants for the same pairs, for example síran 'be afraid'/ (lá-)siran 'frighten', where lá-is the causative prefix.A priori, one might wonder whether there is a semantic distinction between a causative verb such as lá-siran and the transitive use of the corresponding ambitransitive verb síran, but Dumestre's (2011) Bambara dictionary provides examples for which this is unquestionably a case of free variation, at least in the standard variety of Bambara. 16

Conclusion
In the sample of 30 sub-Saharan languages investigated in the first part of this article, the variation in the coding of the noncausal/causal pairs whose noncausal member is a monovalent verb typically selecting an inanimate argument (Questionnaire A) is broadly comparable to that observed for the same pairs (or their near equivalents) in the sample of 21 mainly Eurasian languages analyzed by Haspelmath (1993), with, however, some interesting contrasts.Overall, the ratio of decausative to causative pairs is very similar in the two samples, but the ratio of undirected to directed pairs is much higher in the sub-Saharan sample, within which ambitransitivity ranks first, whereas it ranks fourth in the Haspelmath sample.The sub-Saharan sample includes several languages with an extreme degree of preference for ambitransitivity but no language with an extreme degree of preference for causativization.
As regards the ranking of the pairs of verb meanings included in the questionnaire according to the ratio of decausative to causative pairs, the results are broadly similar to those provided by Haspelmath (1993).By contrast, the very low ratio of undirected to directed pairs for 'dry (intr./tr.)' and the very high ratio of undirected to directed pairs for 'go out/put out (fire)' observed in the sub-Saharan corpus have no equivalent in the Haspelmath sample.
As regards possible typological correlations, the data is consistent with the hypothesis that ambitransitivity and double derivation are incompatible (Nichols et al. 2004: 165) and with the hypothesis that high morphological complexity favors decausativization (NIchols et al. 2004: 166).By contrast, the data contradicts the hypothesis of a correlation between causativization prominence for inanimate pairs and the lack of A-removing/demoting processes (Nichols et al. 2004: 172).
As regards possible relationships between the coding of noncausal/causal pairs, the genetic affiliation of languages and their contact history, the data is not sufficient to draw general conclusions.However, they illustrate the diversity in the coding of inanimate noncausal/causal pairs in two language families (Atlantic and Mande), suggesting the possibility of relatively abrupt changes in the history of languages, since the sample includes two pairs of languages with very different profiles with respect to valency orientation in spite of their relatively close genetic relationship: Seereer/Fula (Atlantic) and Mandinka/Bambara (Mande).
In the second part of the article, I have analyzed the extension of the ambitransitivity strategy to other semantic types of verb pairs in the ten languages of the sample in which ambitransitivity is very strongly prevalent for the noncausal/causal pairs whose noncausal members are monovalent verbs typically selecting inanimate arguments: noncausal/causal verb pairs whose noncausal member is a monovalent verb referring to a process typically involving non-volitional animate entities; noncausal/causal pairs whose noncausal member is a monovalent verb referring to a process typically involving volitional animate entities; pairs of psych verb differing in the perspectivization of the argument-stimulus relationship.
Out of the ten languages that have a very strong prevalence of the ambitransitivity strategy for the semantic type of verb pairs examined in the first part of the article, only three have a sizeable pro-16.This variation in Standard Bambara can probably be analyzed as reflecting dialectal fluctuations in the use of causative derivation, since according to Gérard Dumestre (p.c.), the use of causative derivation is very limited in some Bambara varieties and more common in some others.The free variation observed in the standard variety is probably at least partly due to the fact that Standard Bambara has undergone a strong influence from the Maninka varieties spoken in western Mali, which are closely related to Bambara but in which causativization is more productive than in most Bambara varieties.
portion of ambitransitive pairs for the three semantic types of verb pairs examined in Section 6.This confirms the conclusion that, as suggested by the comparison between English and Basque, a strong prevalence of the ambitransitivity strategy for noncausal/causal verb pairs whose noncausal member is a monovalent verb referring to a process typically undergone by inanimate entities does not necessarily correlate with a high degree of productivity of the same strategy for other semantic types of verb pairs commonly coded as decausative or causative pairs cross-linguistically.
To conclude, I would like to briefly discuss the question of the relevance of Haspelmath's (2016) spontaneity scale for the explanation of the regularities observed in this article.Since the notion of spontaneity scale has been elaborated to account for the distribution of causativization and decausativization strategies across the verbal lexicon, whereas the focus of the present article is on ambitransitivity, the question that arises is whether the spontaneity scale also provides an explanation for the distribution of the ambitransitivity strategy.
The data analyzed in this article suggests that the ambitransitivity strategy is typically used for the segment of the spontaneity scale designated by Haspelmath as "unaccusative", which includes the semantic type of noncausal/causal pairs analyzed in the first part of this article.However, Haspelmath introduces a further distinction between two subtypes of unacccusatives differing in their preferences for the causativization and decausativization strategies: "automatic unaccusative" and "costly unaccusative".Consequently, the question that arises is whether this distinction would also be relevant for explaining the distribution of ambitransitivity across the lexicon.
In fact, the data analyzed in this article shows no obvious relation between Haspelmath's distinction between "automatic unaccusative" and "costly unaccusative" and the variation in the use of the ambitransitivity strategy by individual pairs of verb meanings as described in Section 5.3.In particular, 'dry', which shows the weakest preference for the ambitransitivity strategy, and 'go out (fire)', which shows a relatively strong preference for ambitransitivity', are equally characterized by Haspelmath as "automatic unaccusative".
All this suggests that ambitransitivity would deserve more attention in the cross-linguistic and typological investigation of valency orientation, which has so far focused on causativization and decausativization.Undoubtedly, the languages of sub-Saharan Africa have an important contribution to make to this question.

Abbreviations
11.For a finer-grained analysis of the relationship between causativization and the other strategies in Bantu, readers are referred to Dom et al. (this issue), Laine et al. (this issue) and Yoneda (this issue).
Jóola FóoñiWeelo 'melt' is a strictly intransitive verb that does not lend itself to causative derivation, but nevertheless can be found in an analytical causative construction in which it combines with ekaan 'make' in the role of causative auxiliary.SwahiliNote that the characterization retained here for several Swahili noncausal/causal pairs is controversial, since it is not always easy to determine to what extent their history should still be considered relevant in a synchronic analysis.See Dom et al. (this issue) for a discussion of alternative analyses.TswanaIn ša/fisa 'burn (intr./tr.)', the causative suffix -is is clearly recognizable, and historically, this is undoubtedly a causative pair.Synchronically, it can still be analyzed as a causative pair with stem allomorphy triggered by the causative suffix, although an analysis in terms of suppletion could also be considered.(intr./tr.)cà̰ rî, yὲrὲkὲ/cà̰ rî, yὲrὲkὲ amb 13. turn upside down (intr./tr.)wá-kà̰ ŋa̰ /wá-kà̰ ŋa̰ amb Baule Baule has no causal counterpart of wu 'dry (intr.)',but an analytical causative construction is possible (Jérémie Kouadio, p.c.).upside down (intr./tr.)kpɛ ... wun/kpɛ ... wun amb upside down (intr./tr.)kálákktəkin/kálánggin dec Sandawe For pairs 12 and 13, I was not able to find the exact equivalents of 'spread' and 'turn upside down', and I took the liberty of replacing them by semantically close verbs.upside down (intr./tr.)tɬ'êʔe̥ /tɬ'èʔe̥ sé (deviate) cau LingalaIn pairs 9, 10 and 13, the stem ends with the reversive suffix -ol, and -wan is the realization of the underlying sequence -ol-an.