Do Linguistic Properties Influence Expressive Potential? The Case of Two Australian Diminutives (Gunwinyguan Family)

Abstract:Although expressivity is indisputably a crucial function of language, expressive features have often been neglected in linguistic descriptions. After discussing how this lack can reflect a gender bias in language documentation, this article recruits first-hand data from two Australian Aboriginal languages to explore whether the grammatical properties of individual languages can influence the semantics of their expressive resources. The study compares diminutives found in Dalabon and Rembarrnga, two Gunwinyguan languages spoken in the same cultural environment. The comparison shows that in spite of many mismatching linguistic properties, the emotions that these diminutives can express remain remarkably stable, suggesting strong sociocultural constraints in this semantic domain.

convergent contexts, mostly on nouns referring to children-as is overwhelmingly common across languages (Wierzbicka 1984;Dressler and Merlini Barbaresi 1994;Jurafsky 1996;Grandi and Körtvélyessy 2015a;Ponsonnet 2018a).However, these two diminutives differ in their etymologies, and a thorough analysis reveals nuances in their morphological status, denotational meanings (size or age), semantic prevalence (i.e., whether they are used primarily to describe size and age, or to express emotions), and morphosyntactic distribution.How do these linguistic differences impact on the expressive functions of these two diminutives?
The main conclusion of the study is that irrespective of other mismatching linguistic properties, the emotions that speakers can express with diminutives are very similar in Dalabon and in Rembarrnga.They are also consistent with the emotions typically encoded by miscellaneous expressive resources in other languages in the same region, and perhaps elsewhere on the Australian continent.Thus, emotional semantic categories seem relatively impervious to variation in linguistic properties such as etymologies, form, denotational semantics and morphosyntactic distribution.We may hypothesize that the sociocultural context shared by Dalabon and Rembarrnga prevails over linguistic differences in this respect.On the other hand, the study also suggests that linguistic properties may impact on the expressive potential of linguistic resources in other ways.In the case at stake, differences in the respective morphological status and morphosyntactic distribution of the two diminutives imply that they are not used in exactly the same situations.This may bring differences in when and what speakers can actually communicate with these tools in their respective languages.To summarize, the emotional semantic categories encoded by expressive tools are relatively independent of the linguistic properties of these tools, but linguistic properties could make a difference as to when and how these semantic categories can be applied in actual communication.
Each section in the article describes and compares the Dalabon and Rembarrnga diminutives with respect to different properties.Etymology and morphological status are discussed together in section 3.Sections 4 and 5 present denotational meanings and nondenotational functions, respectively.Section 6 deals with the morphosyntactic distribution of these diminutives, and section 7 with their respective semantic prevalence (which is presented after the distribution because it depends upon it in Dalabon).
As background to the linguistic analyses of the expressive features, section 2 presents the linguistic context of the study and addresses the methodological issues surrounding the documentation of diminutives and expressive resources in general.I show how gender-related practices and expectations about storytelling and language documentation have historically limited the investigation of the expressive and emotional dimensions of language-in this case, preventing the identification of diminutives and their salient emotional values in a number of Australian languages.
Both in precolonial times and in more recent years, Dalabon and Rembarrnga speakers belong to a relatively unified cultural region that also included the Jawoyn language and eastern Bininj Gun-wok dialects.These language groups intermarry and share social life, narratives, and rituals.This shared cultural background may explain some of the semantic convergences relating to emotions that are discussed in this article.Dalabon and Rembarrnga, in particular, have been in intense contact for a long time-contact that includes widespread multilingualism.Although Dalabon belongs to the Central branch of the family and Rembarrnga to the Eastern branch (figure 1), the two languages share the core of their morphology and grammar, as well as many cognates and lexical distinctions.Like all other Gunwinyguan languages, Dalabon and Rembarrnga are largely head-marking and highly polysynthetic, with some casemarking nominal morphology mainly targeting adjuncts.(For Rembarrnga, see McKay 2011;Saulwick 2003aSaulwick , 2003b; for Dalabon, see, inter alia, Evans, Merlan, and Tukumba 2004;Evans and Merlan 2003;Evans, Brown, and Corbett 2001;Ponsonnet 2015;Evans 2017;Luk and Ponsonnet forthcoming.)Figure 1.Structure of the Gunwinyguan family (Evans 2003:33).

Corpora.
The analyses presented here rely upon firsthand data comprising audio-and video-recorded speech of various genres, both narratives (including personal emotional narratives) and stimuli-based elicitation designed to collect close-to-spontaneous emotional data (Ponsonnet 2014a).Stimuli-based data-collection relied on a range of methods such as pictures designed for elicitation, video clips from Baron-Cohen's (2003) Mind-Reading Library, and a number of mainstream Australian movies with culturally appropriate topics and scenarios. 1As a result of the demographics of remaining speakers as well as local gender dynamics (see section 2.3), data was collected almost exclusively from female speakers in both languages (nevertheless, there is evidence that male speakers can also use diminutives, at least in Dalabon).In the examples used in the present article, translations combine those collected from speakers in Kriol (the English-based creole that has replaced Dalabon; see Ponsonnet (2017Ponsonnet ( , 2018b, forthcoming a) , forthcoming a) with interpretations imposed by the context.
For qualitative analyses, I rely upon extensive corpora of different sizes in each language.My Dalabon corpus in toto is much larger-about sixty hours, mostly collected between 2007 and 2014 from four mature female speakers.The Rembarrnga corpus comprises a bit under nine hours, collected between 2014 and 2016 from three mature female speakers.For quantitative analyses, I used smaller data sets comprising comparable types of data (mostly emotional narratives and comments on movies) in comparable proportions for each language.The Dalabon data set is about ten hours long and contains ninety-eight tokens of the diminutive.The Rembarrnga data set is a bit more than four hours long and contains forty-one tokens.In spite of the disproportion in size, the data sets provided comparable figures when converted to percentages, because their contents are commensurable.

Gender, genres
, and the shortcomings of language documentation.

Gaps in linguistic descriptions.
As in many languages in the world, diminutives are overwhelmingly frequent in Dalabon and Rembarrnga, at least in female speech. 2As an indication, the corpora considered for qualitative analysis in this article contain about one diminutive every six minutes, and comparable figures hold across the rest of my corpora.While this is certainly not representative of every genre or register in either of these languages, the corpora in question are not at all limited to children-oriented speech, for instance.Instead, as presented in section 2.2, they contain a broader range of narratives and descriptions which are probably somewhat comparable to the contents of every day conversations in this cultural context.
Despite their frequency, diminutives have received very little attention in previous descriptions of Dalabon and Rembarrnga-something that is all the more surprising given that these descriptions are extensive and thorough.Prior to Ponsonnet (2014b), the literature on Dalabon (a full dictionary, two doctoral dissertations, and half a dozen articles; see section 2.1) made no mention of diminutives; in fact, diminutives were believed to be absent (Evans p.c. 2012).As for Rembarrnga, Saulwick (2003a) did mention the diminutive form in his dictionary, but labeled it an adjective; he then discussed it in his chapter on evaluative forms, with reference to the recently identified Dalabon diminutives (2015).McKay's (2011) Rembarrnga grammar, in spite of its breadth and precision, only mentions the diminutive very cursorily (a cell in a table, 2011:92).None of these publications identify emotional functions.Ponsonnet and Evans (2015:406) recognize that diminutives had long been omitted in descriptions of Australian Indigenous languages-Heath, for instance, writes that "Aboriginal languages are generally weak in diminutives and other hypocoristics" (2015: 226;Grandi and Körtvélyessy 2015a:401-52).
Owing to the level of endangerment of Dalabon and Rembarrnga and the demographics of their speakers, I documented these languages mainly with women over sixty years of age (section 2.2); Heath (2015) suggests that this may explain why diminutives appear disproportionately frequently in my data.However, middle-aged female speakers of another neighboring Gunwinyguan language, Bininj Gun-wok (Central branch), also used diminutives about as frequently as my Dalabon and Rembarrnga consultants, or even more so.The diminutive in the Kunwinkju dialect of Bininj Gun-wok is the verbal prefix yaw¤, which is also a diminutive in Dalabon (Ponsonnet 2014b:81).The form yaw¤ can be an incorporable noun that means 'small one' (Evans 2003:473), but example (1) (Kunwinjku dialect) shows that yaw¤ is also a prefix that expresses compassion towards an adult (who is not small)-an emotional extension typical of Gunwinyguan diminutives (see section 5).
( 1 In the Kune dialect of Bininj Gun-wok, seen in ( 2), speakers use the form =wurd, which is also the most frequent diminutive in Dalabon (see Rose 2018 on diminutives as dialectal markers).
(Kune) 3SG>3SG.H-hit-PPFV=DIM 3SG>3SG.H-cheek-smash-PPFV DEM 'He hit her, poor thing, he smashed her cheek there.' (BGW_20170521_002_MM 144 [Stim]) [about a picture of an adult woman being hit by her husband] In spite of their high frequency in (at least) these dialects, Evans's (2003) grammar of over seven hundred pages, a model linguistic description in many other respects, makes no mention of diminutives.In the light of their prevalence in four Gunwinyguan languages or dialects across two different branches of the family, 3 it is puzzling that these diminutives had received so little attention in previous descriptions of these languages.Even where extensive grammatical discussion may have been deemed unnecessary, diminutives are a significant linguistic feature, given their frequency, that one would think ought to be presented when accounting for the languages' grammatical resources.In the following sections I discuss how this blind spot, which reflects a more general lack of attention to expressivity in grammatical descriptions, may result from certain biases in language documentation practices, and, in particular, some gender biases.

Documentation and gender.
It is well known that linguistic elicitation oriented towards grammatical description tends to produce very unnatural linguistic data, which is likely to exclude spontaneous emotional features.However, grammatical elicitation is normally combined with the collection of texts (Chelliah 2001), and indeed the appendices of McKay's (2011) grammar of Rembarrnga and of Evans's (2003) grammar of Bininj Gun-wok contain sizeable collections of very rich narratives.Why did these narratives fail to reveal the prevalence of diminutives?
In my data, diminutives are used by female speakers when recounting emotionally loaded episodes of their personal lives (including hardships, love stories, deaths, health matters, etc.), commenting on interpersonal relationships and other people's behaviors both in dramatic contexts (e.g., movies) and in ordinary contexts (pictures, clips, everyday life anecdotes), or telling stories for children, for instance.Although I have occasionally heard men using diminutives, these contexts are more typical of female speech.Such contexts are probably relatively common themes in everyday conversations, but they rarely feature among the texts appended to grammars.
The narratives in McKay (2011) and Evans (2003) were mostly recorded from male speakers, with the exception of one short narrative by a female speaker (Evans 2003:712).This asymmetry has been prevalent in the documentation of Australian languages.The more than 550 pages of Heath's (1980) volume of Wubuy (Nunggubuyu) texts (a companion to Heath's 1984 grammar of this language, another masterpiece of linguistic description), contain an impressive collection of 170 narratives authored by seven male speakers, but just a single narrative by a female speaker (Heath 1980). 4 Australian societies are structured around a strong gender division, which makes it difficult for men to record women.However, the reverse is less true: females often record males.For instance, Glass and Hackett (1979), a volume edited by two female authors, contains fifteen Ngaanyatjarra texts (Western Desert language, Pama-Nyungan family)-thirteen by a man and two by a woman.Of course, the gender imbalance is not always so marked, and patterns are evolving.The tendency for linguists to collect data from male speakers (see also section 2.3.3) has been mitigated in recent years.This is both because there are now more female linguists doing fieldwork (like myself), and because female speakers, who are often among the "last speakers" of their language, have been able to express their interest and expertise.For instance, Merlan's (2016) collection of thirty-six Jawoyn narratives (Gunwinyguan) collected throughout the 1980s, 1990s, and early 2000s contains twenty-four pieces by three women and only eleven by one male contributor. 5

Gender and Australian
Aboriginal "folk stories."Another reason why narratives collected in ordinary language documentation do not adequately represent expressive features such as diminutives is the genre or theme of the stories collected in such language documentation.McKay's (2011) and Evans's (2003) grammars contain "bush adventures" narratives (about fishing, hunting, mastering cattle, etc.) and mythological stories (about animal ancestors, spirits, ceremonies).The etiquette of language documentation tends to be highly codified in Australian communities, so that it is often difficult to record outside of a conventional repertoire of "folk narratives" (Carew 2016;Walsh 2016).Apart from hunting adventures or mythological stories, other common themes are ceremonies, kinship and social rules, arts (songs and paintings), traditional techniques, ethnobiological knowledge, and colonial history (Hercus and Sutton 1986;Charola and Meakins 2016).These themes do not necessarily reflect the full spectrum of daily preoccupations and discourse in Australian Indigenous communities, and the corresponding folk narratives are also stylistically marked (Carew 2016).While a more extensive and systematic study would be necessary to reveal how the range of genres represented in ordinary linguistic documentation correlates with the grammatical features identified in the resulting linguistic descriptions, the bias in coverage in such documentation is readily evident even from a cursory survey.
The nature of this language-documentation repertoire is, in fact, not independent of the gender divisions alluded to in section 2.3.2.In "traditional" Australian Aboriginal societies, each gender was assigned a fixed specialization in terms of daily tasks, expertise, and knowledge, whether mundane or ritual (Berndt 1950(Berndt , 1974;;Merlan 1992;Ponsonnet 2018c); for example, hunting was a male task, while gathering vegetable food was a female task.In many parts of the Top End (where Gunwinyguan languages are spoken), men controlled the most significant ritual and mythological knowledge (Cowlishaw 1979;Keen 1994) and conducted prestigious activities, such as painting and singing.The female domains of expertise were domestic, including, in particular, anything related to reproduction and raising children (Cowlishaw 1978).These domains were canonically nonprestigious, even vulgar, shameful, and indeed taboo.Among the Dalabon and across this region, women were expected to protect men from being exposed to any matters related to sexuality and reproduction (Maddock 1970;Ponsonnet 2014bPonsonnet :216, 2018c).As a result, female matters are not considered suitable topics for discussion with outsiders and are not part of the folk repertoire of narratives that qualifies for recording with researchers.This helps to explain why it is easier to collect narratives from men in Australian languages: although there are female storytellers talking to outsiders, in general and by default, this activity often falls within the male domain because it is prestigious.
In line with these observations, some women I work with on elicitation tasks have declined to record narratives themselves, deferring to their husband on the grounds that "he has all the good stories."When women do record narratives, some of them choose from the male-oriented authorized repertoire: the mythological narratives that they know, female techniques such as basket weaving, or malelike stories such as fishing adventures 6 (e.g., the female narrative in Evans [2003:712] is entitled "Getting Crocodile Eggs").Kinship, ethnobiology and colonial narratives, which are unmarked for gender, are also possible topics.But typically female aspects of life-for instance, household organization, children's education, or child-bearing-although indisputably important, are not considered acceptable topics a priori.In fact, child-bearing was sometimes explicitly discarded as inappropriate for documentation.This excludes a lot of the contexts where typically female expressive features such as diminutives are likely to occur.Linguists are thus presented with a dilemma.On the one hand, respecting the established local repertoire has evident ethical benefits, such as remaining neutral with respect to local social structure and faithful to speakers' inclinations.On the other hand, encouraging speakers to step out of their repertoires also has various benefits: giving a voice to female speakers within these groups, covering a broader range of themes, including some closer to daily domestic life, and documenting additional speech registers.
A final point to consider is that the difficulty of recording females or female themes does not result only from the local social structures.Instead, several generations of researchers across disciplines have maintained an emphatic interest in male repertoires-mythological and ceremonial matters, as well as painting and singing in particular-thus reinforcing the local distribution of prestige.In my experience, female speakers in the Gunwinyguan region are quite inclined to incorporate new themes into their repertoire of "appropriate narratives" when given enough time and opportunities to do so (Ponsonnet 2014a). 7Independently of linguistic documentation, newly established female genres have also emerged, such as personal introspection in the context of Christian Fellowship groups, which speakers are inclined to transfer to the context of language documentation.Finally, several of my female consultants seemed eager to record miscellaneous personal matters, sometimes explicitly characterizing the documentation context as an opportunity to talk about things that they could not discuss with their family.These observations suggest that the folk-codified repertoire is actually relatively flexible and, in turn, raise the question whether the apparent inflexibility of this repertoire might in part reflect the expectations of the white audience.
It is difficult to tease apart the respective roles of local gender structure and narrative practices, on the one hand, and of pressures perpetuated by white interlocutors (including researchers), on the other hand.In any case, genderbiased judgments of prestige imposed on genres and themes in language documentation have been a strong impediment to the understanding of many aspects of Australian languages, the encoding of emotions being a typical example.In the case of Gunwinyguan diminutives, this has resulted in incomplete grammatical descriptions.In order to collect data that is more representative of a diversity of semantic domains, linguists should actively seek ways out of the patterns described above (see Ponsonnet [2014a] for practical suggestions).
After this methodological preamble, I turn to the description and comparison of the diminutives in Dalabon and Rembarrnga.

Form, etymology, and morphological status.
3.1.Dalabon =wurd.Dalabon diminutives are discussed in detail by Ponsonnet (2014b:81-109; see also Ponsonnet and Evans 2015).The sections below present a summary of this discussion, emphasizing the points that are relevant for the comparison with the Rembarrnga diminutive.The most frequent Dalabon diminutive is =wurd [wuÔ], 8 which is segmentally homophonous with the noun wurd 'woman's child' and related to the current noun for child, wurdurd. 9This conforms with Jurafsky's (1996:538) findings that 'child' is among the most common etymological sources for diminutives cross-linguistically.As illustrated in (3), wurdurd 'child' often combines with the diminutive =wurd.
Ngarrkun¤no.3SG-R-cry:REDUP-PRS newborn.baby-FILL'And where's his mother?This little child is crying.This newborn baby.' (20120705b_004_MT 086 [Film]) [about a toddler who has been taken away from their parents and placed in an institution] The form is better seen as an enclitic 10 given its broad distribution (section 6.1) and its prosodic and phonological integration with its hosts.The initial consonant in =wurd is often pronounced as a voiced pharyngeal approximant (represented for convenience as <H> in Dalabon), especially after vowels and nasal consonants.

Rembarrnga (¤)kanja(ng)h.
The diminutives in Dalabon and Rembarrnga are both postposed: they follow their host.However, they vary in form, etymology, and morphological status.The Rembarrnga form, (¤)kanja(ng)h [gaça(×)ý] 11 (Saulwick 2015), attracts a range of possible realizations, as the final coda can be reduced to a glottal stop, or even deleted altogether.For some speakers, [gaça×ý] is the most frequent realization, while others predominantly use [gaçaý].Some realizations of the form can be regarded as suffixes (especially when reduced, ¤kanja(h)) occurring between the host and other modifiers (adjectives, possessives, case markers), with a high degree of prosodic integration, as in (5).Alternatively, the form can also occasionally be prosodically independent, especially with the longer realization kanjangh. 12However, even when realized as an independent form, the diminutive cannot function alone; it depends syntactically upon the preceding element (section 7).
( 5 Dalabon =wurd and Rembarrnga (¤)kanja(ng)h differ not only in form, but also in etymology.While in Dalabon the transparent etymological source of =wurd is a noun meaning 'child', in Rembarrnga the standard noun for 'child' is dakku, which is evidently not a candidate source for the Rembarrnga diminutive (¤)kanja(ng)h. 13The etymology of (¤)kanja(ng)h is most probably an adjectival form meaning 'small'.Indeed, Saulwick (2003a:29) describes kanjarngh as an adjective ('small'), and a cognate adjective kanjah [gaçaý] 14 'little' is reported in Ngalakgan, another Eastern Gunwinyguan language to the southwest of Rembarrnga (Merlan 1983:195).On the western side of the Eastern Gunwinyguan branch, a cognate diminutive form is reported in Ngandi (suffix [gaçaý]).Warndarang, of the neighboring Marran family (southeast), and Ritharrngu, a Yolngu language neighboring Rembarrnga and Ngandi on the northeast, both have cognate suffixes (respectively [gaça] and [gaça×ý]/ [×aça×ý] [Heath 1978:145]).The form is thus originally an adjectival root expressing small size that has lost syntactic and morphological independence to become a diminutive suffix in the easternmost languages of this region, and is in the course of doing so in the westernmost languages.

Denotational semantics.
As with many diminutives across the world's languages (see, e.g., Dressler and Merlini Barbaresi 1994;Grandi and Körtvélyessy 2015a;Ponsonnet 2014b:81-109), the semantic extensions of Dalabon and Rembarrnga diminutives can be divided into three broad types: denotational meanings, having to do with quantitative or scalar evaluation; emotional meanings or connotations, when the use of a diminutive relates to the emotional attitude of the speaker at the time of speech; and pragmatic effects, when the diminutive mitigates the negative impact of a statement, for instance with a politeness effect.These three types are not exclusive of one another-emotional connotations are often embedded in denotational senses (see Ponsonnet [2014b: 81-109] for a discussion)-but they are conceptually distinct, and the present section deals with denotative meanings only.
At first sight, =wurd in Dalabon and (¤)kanja(ng)h in Rembarrnga occur in comparable contexts, namely, referring to young humans.These usages are both rendered as lil (from English little) in speakers' Kriol translations for both languages.However, probably due to the distinct etymologies of these two diminutives, their respective denotational meanings are underlyingly different, with the Dalabon diminutive targeting young referents and the Rembarrnga diminutive targeting small referents.

Dalabon =wurd.
In many languages the denotational meaning of diminutives is 'small', but in Dalabon =wurd means 'young', at least when it is affixed to a human noun, which is the most frequent case (see section 6.1).This is illustrated in (6).Indeed, =wurd often qualifies humans that are not small as long as they are young, as illustrated in (7), where the referent is a relatively tall and strong young man in his early twenties (see figure 2).Conversely, throughout the Dalabon corpus, diminutives were never used on nouns referring to humans who are not young.On nonhuman nouns, =wurd can mean 'small' (with animals, body parts, and inanimate count nouns) or 'a little bit' (with mass nouns), as illustrated in ( 8) and ( 9).As discussed in section 6.1, =wurd is overwhelmingly less frequent on nonhuman nouns, though, and therefore the denotational meaning 'small' of =wurd is secondary compared to the denotational meaning 'young'.(8) (10), where it qualifies a child who is about twelve years old (see figure 3).In Dalabon, the diminutive also very frequently qualifies young adults, even if they are not small in stature-the denotational meaning is 'young' (section 4.1).But this does not occur in Rembarrnga: in my corpus, (¤)kanja(ng)h is never attested on nouns referring to young adults, although speakers had many opportunities to use it in this way.On the other hand, it is attested with adults who are small in stature. 16Given that (¤)kanja(ng)h etymologically means 'small', it is expected that this is also what the diminutive enclitic means denotationally.This explains why it applies to children as opposed to young adults, as they are actually small in size, while young adults are not usually smaller than older adults.
When (¤)kanja(ng)h is encliticized to nonhuman referents, the meaning is also 'small', or 'a little bit' on mass nouns, just like in Dalabon.(11)  the Rembarrnga (¤)kanja(ng)h means 'small'-a nuance that could reflect the respective etymologies of these forms.Thus, in these two closely related languages spoken within a unified cultural context, two comparable linguistic tools can differ semantically because they do not share the same etymological origin.

Nondenotational functions: emotional and pragmatic values.
Emotional meanings and pragmatic functions (see section 4) are grouped together in this section as "nondenotational" usage, since both are less referential and more subjective; moreover, pragmatic usages are usually tinted with emotions (affection in particular).Nondenotational semantics is very consistent across the two languages, with Dalabon and Rembarrnga diminutives essentially expressing the same emotional values.Pragmatic functions are less frequent in my Rembarrnga data than in my Dalabon data, but this may be a bias of my corpora.

Emotional meanings of =wurd.
As is the case for many diminutives across the world (Ponsonnet 2018a), Dalabon =wurd is used to express emotional states of speakers, mostly with respect to three types of positive emotions, which can be summarized as follows: • compassion and related notions: expressing compassion for someone or expressing approval and endearment when witnessing compassion ("seconddegree compassion"); • affection; • endearment when witnessing personal, intimate routines.These emotions are not mutually exclusive, and they often apply at the same time for a single occurrence of the Dalabon diminutive.However, they are conceptually distinct and worth stating as such because each of them can, of itself and independently of the others, trigger the use of a diminutive.In many cases, emotional values also co-exist with denotational meanings (as discussed in section 7).
Expressing compassion, or approval and endearment when witnessing compassion was the most frequent nondenotational function of =wurd in the corpus I used for this study.(On the expression of compassion by diminutives crosslinguistically, see Ponsonnet [2018a:27].)The Dalabon diminutive can express compassion for someone going through hardship (see example (3) in section 3.1), or the speaker's approval and endearment when witnessing people being compassionate with each other.This applies in particular when witnessing people expressing attachment or grief for others, taking care of each other, as in (12), or sharing material goods (which in this cultural context is an expression of affection [Ponsonnet 2014b:196-99;Myers 1979]).In these contexts, diminutives (in Dalabon as in Rembarrnga, see section 5.3) are often translated in Kriol using the compassionate interjection bobala (from English 'poor fellow').Compassion towards oneself, on the other hand, is not clearly attested with =wurd.The expression of affection for someone is also a common emotional sense of =wurd.Affection is the most commonly attested emotional extension for diminutives across the world (Ponsonnet 2018a:23-24), but in Dalabon, affection is less frequent than compassion.In (7) in section 4.1 above, the speaker is commenting on the movie Ten Canoes (see n. 1), and she refers to a character that she is particularly fond of.Throughout the recording sessions, her references to this character used a diminutive by default, in contrast with other characters that she was more critical of (see Ponsonnet [2014a:87-92] for a detailed discussion).
The last type of emotional context where diminutives are typically used is when talking about intimate routines (for the typological status of this context, see Ponsonnet 2018a:32-33), i.e., daily habits that one enjoys on one's own, such as listening to music before going to sleep, combing one's hair softly and with care, or sitting by a fire alone in the dark, as in (13)-a comment on the scene from the movie Samson and Delilah (see n. 1) shown in figure 4.
(13) Mimal ka¤h¤marnu¤rurung 17   All these three types of emotional contexts are attested for diminutives in other languages in the world (Ponsonnet 2018a), but affection is often more frequent than compassion, whereas as mentioned above, in Dalabon it is compassion that is more frequent.Unlike many diminutives around the world, Dalabon diminutives do not express negative emotions such as contempt (Ponsonnet 2018a:25).Data recently collected on other languages in the same geographical area suggest that this particular threefold combination of emotional meaningscompassion, affection, intimacy in personal routines-is characteristic of expressive resources in this part of Australia.That is, expressive features such as interjections or prosodic contours typically display this threefold set of values, in Dalabon as well as in other languages spoken in this area-Rembarrnga, Bininj Gun-wok dialects, and Kriol (Ponsonnet forthcoming a).Given the degree of resemblances between social groups in Aboriginal Australia with respect to emotions (Ponsonnet 2016, forthcoming a), as well as with respect to other domains, such as cosmologies or kinship (Dousset 2012) for instance, it seems possible that the semantic convergence of expressive categories may extend beyond the Gunwinyguan region.

Pragmatic functions of =wurd.
Finally, and again like many diminutives across the world (see, e.g., Dressler and Merlini Barbaresi 1994), the Dalabon diminutive =wurd can be used for pragmatic attenuation.This is typically for minimizing the benefits received by the speaker (as in ( 9) in section 4.1 above), the degree of imposition implied by a request (as in 'give me a little bit of . ..'), or the negative aspect of a situation, as in ( 14).The diminutive =wurd is also marginally attested for self-modesty (by a male speaker).( 14

Pragmatic effects of (¤)kanja(ng)h.
Like the Dalabon diminutive =wurd, (¤)kanja(ng)h can be used to minimize the negative impact of a situation.In (18), the speaker is correcting the description of a stimulus by an older and more qualified speaker who is her classificatory aunt.Given the tone of deference and respect that the speaker of (18) used towards the senior speaker throughout our language sessions, endearment is not a likely interpretation for the diminutive in this example.Instead, as confirmed by the intonation, the diminutive attenuates the potential impoliteness of the junior speaker's correction.Such pragmatic usage of the diminutive was rarer in my Rembarrnga corpus than in my Dalabon corpus.Although Dalabon =wurd and Rembarrnga (¤)kanja(ng)h differ in their denotational semantics (section 4), the contrasts are not carried over to the emotional values of these diminutives.Thus, the respective etymologies of these linguistic devices seem to affect their denotational meanings, but not their expressive values.

Dalabon =wurd.
In Dalabon, =wurd occurs on a very wide range of hosts, including at least nouns, verb complexes, adjectives, pronouns, demonstratives, numerals, adverbs, and interjections.In other words, it seems that most word classes can receive a diminutive, given an adequate semantic context.However, nominals (including adjectives and demonstratives) and verbal compounds are by far the most frequent bases for the diminutive; between them, they account for 98 percent of all occurrences in the corpus under consideration.
Nouns (excluding other nominals) account for 75 percent of all occurrences, and among them, nouns with a human referent represent 92 percent (or 69 percent of all occurrences).Diminutivized nouns are typically nouns denoting gender and stages of life, e.g., kirdikird 'woman (female human)', yawurrinj 'young man', as in ( 6) and ( 7) above (section 4.1), as well as kin terms, as in ( 19).Other nouns referring to humans, such as nes=wurd 'the young nurse', also occur, albeit more rarely.Verb complexes are the next most frequent category, as illustrated, for instance, in ( 14) in section 5.2 above, and in (20).The proportion of occurrences of =wurd on verb complexes is much smaller than on nouns, but nevertheless significant: 17 percent of all occurrences.The respective prevalence of each of the meanings discussed in previous sections (denotational vs. nondenotational meanings) varies with the nature of the base, and this is further discussed in section 7.

Rembarrnga (¤)kanja(ng)h.
An important difference between the Dalabon and Rembarrnga diminutives is that in Rembarrnga, occurrences on verbs are marginal (three occurrences, 7 percent), and, in fact, are not attested with a clearly suffixed form.In ( 21), for instance, loudness drastically decreases at the end of the second verb complex, yarra¤bba¤nini, giving the impression of a pause before kanjangh, which bears clear primary stress on its first syllable (and lengthening on the second syllable).(In this example, stress on the following syllable is marked by the IPA stress marks: ' for primary stress, and / for secondary stress.)The meaning of all three postverbal occurrences is emotional (see section 7.5), and overall, the free postverbal diminutive kanja(ng)h is evocative of a type of compassionate interjection found in many Australian languages (which originate etymologically in nominals and are often translated as 'poor fellow', Ponsonnet forthcoming b).In the Gunwinyguan region, for instance, Dalabon has weh-no, 20 illustrated in (22), which is also an adjective meaning 'bad', and Kriol has bobala (from English 'poor fellow'), which often serves to translate diminutives (see Ponsonnet forthcoming a: chap.6).While it is not uncommon for these interjections to follow a verb, they can also occur anywhere else in the clause (typically clause-initially), or alone in single-word utterances.Diminutives do not occur in this range of contexts, but one might wonder if kanja(ng)h could evolve in this direction, given its morphological autonomy when following verb complexes.As illustrated in (10) in section 3.2, ( 15) in section 5.3, or (18) in section 5.4, nouns referring to humans (gender, stages of life, kin terms) are the most common hosts in Rembarrnga, as in Dalabon.Yet, the proportions differ.In Dalabon, nouns with a human referent represent 92 percent of the occurrences of =wurd on nouns, but this figure only amounts to 79 percent for (¤)kanja(ng)h in Rembarrnga, where 21 percent of the nouns refer to inanimate entities (compared to only 5 percent in Dalabon).This could be explained by the fact that Rembarrnga (¤)kanja(ng)h targets small referents and essentially applies denotationally to children, while the Dalabon =wurd applies to children and young adults.The sense 'small' also implies that (¤)kanja(ng)h can qualify adults who are small in stature albeit not young, but this is rare, perhaps because adults are not usually endearing.The Rembarrnga diminutive is also proportionally more frequent on parts of speech that refer to humans but are not nouns, such as numerals or demonstratives for instance (as illustrated in ( 16) in section 5.3 above).Such alternative parts of speech represent 16 percent of the occurrences of (¤)kanja(ng)h on noun phrases, while in Dalabon this only amounts to 7 percent.To summarize, the predominance of nouns referring to humans is less clear in Rembarrnga, where the diminutive occurs relatively often on nouns that refer to nonhumans, as well as on parts of speech other than nouns.

Summary.
In Dalabon, as in Rembarrnga, diminutives are fairly ubiquitous on nominals and most frequent on nouns.However, while the prevalence of human nouns is overwhelming with =wurd in Dalabon, in Rembarrnga (¤)kanja(ng)h displays a more balanced profile.The languages also differ with respect to verb complexes; while these are common hosts for the Dalabon diminutives, they are only marginal in Rembarrnga, where verbal occurrences appear to be free forms akin to interjections.Thus, in spite of the grammatical resemblances between the two languages, the morphosyntactic behaviors of the diminutives diverge.

Semantic prevalence.
It is a well-attested property of diminutives that their semantics is modulated by the morphosyntactic class of their host (Bauer 1997).This is certainly the case in Dalabon, as investigated in sections 7.1-7.4,but does not apply in Rembarrnga, where diminutives are always primarily emotional (section 7.5).

7.1.
Dalabon =wurd on nouns referring to humans.I explained in section 4.1 that the denotational sense of =wurd when encliticized on a noun referring to a human being is 'young' (not 'small').This meaning is always active; in my corpus, =wurd does not occur on nouns referring to humans who are not young.As is discussed in section 7.3, compassion for older persons is often expressed with a diminutive on the verb complex heading the clause.
With human nouns, emotional extensions are better treated as connotations, since they do not occur independently of a denotational sense.The evidence for these emotional values is then quantitative; the diminutive may occur in emotionally neutral contexts, but the quantitative analysis of its distribution shows that the emotional contexts listed in section 5 (compassion, affection, and endearment when witnessing intimate routines) strongly favor the use of diminutives on nouns referring to young humans (Ponsonnet 2014b:81-109).

Dalabon =wurd on other nouns.
As also pointed out in section 4.1, the denotational meaning of =wurd can be 'small' (as opposed to 'young') when the base noun refers to nonhuman entities: animals, body parts, inanimate things (count or mass nouns).With inanimate nouns, the denotational sense can sometimes be barred in favor of a purely emotional meaning.In (23), for instance, =wurd modifies blankid 'blanket'.The sentence is a comment on the movie Rabbit-Proof Fence (see n. 1), where three children heroines are being chased by nasty people.In this scene, they hide under a blanket to escape their pursuers.Here the denotational sense 'small' is inactive.The blanket can hardly be described as small, since it is large enough to cover the three children entirely.In spite of the construction, the blanket is not topical in this statement; the speaker does not actually intend to say something about the blanket, since she is, in fact, talking about the children's dire situation.As confirmed by the intonation, =wurd only has emotional functions here-to index the speaker's compassion for the children who are at risk of being caught, as well as her endearment at their solidarity as they hide together.Hence, with inanimate nouns, contrary to nouns referring to humans, the denotational sense is optional and the emotional sense can occur alone.However, occurrences of the diminutive on nouns referring to nonhuman nouns are a minority: only about 5 percent of the total number of occurrences in the corpus under consideration (section 6.1).

Dalabon =wurd on verb complexes.
With verb complexes also, the order of prevalence of the different types of meanings is inverted compared to human nouns; denotational meanings are optional, and emotional meanings are always active and can occur alone.Many occurrences may be interpreted under one of the denotational senses: one of the arguments is young or small, as in (24) (repeated from (20) for convenience), or the event is incomplete (as in ( 14) in section 5.2 above) or lacks intensity.
(24) Woywoy bulu ka¤h¤naHna¤n=wurd bulu (Dalabon) INTJ.COMPASS 3PL 3SG>3-R-see:REDUP=DIM 3PL ka¤h¤djukkodjukko¤n. 3SG>3-R-wash:REDUP-PRS 'Good on her, she kindly looks after them (the children), she washes them.' (20120705b_004_MT 025 [Film]) [a child is taking a shower with the help of one of her caretakers] However, there are occurrences where none of these denotational meanings applies.In ( 25), the speaker comments on the movie Ten Canoes (see n. 1) where the hero falls over, badly wounded.The hero (see figure 5) is a brave, mature man who can hardly be described as young or small in stature.The fall cannot be described as incomplete (the man is lying down) or lacking in intensity, as it is quite brutal.Here, the diminutive =wurd solely encodes the speaker's compassion about the hero of the movie, independent of any age or size assessment, or incompleteness of the event.Hence, there is no denotational meaning, but only an emotional meaning.diminutive =wurd shifts when it is encliticized on a nonhuman noun or a verb complex, as opposed to a human noun.With human nouns, the denotational meaning is prevalent; it is always present, and nondenotational values (emotional or pragmatic) are better described as connotations.With nonhuman nouns and verb complexes, nondenotational meanings are prevalent; they are always present, while denotational meanings are optional.Hence, in Dalabon, as in many languages, the productivity and the semantic prevalence of diminutives is influenced by the morphosyntactic status and the semantics of the base (see Bauer 1997:540, following Nieuwenhuis 1985:221-23). 217.5.Rembarrnga (¤)kanja(ng)h.Contrary to Dalabon, in Rembarrnga the denotational meaning 'small' of (¤)kanja(ng)h is always optional, including when the host is a human noun.Indeed, (¤)kanja(ng)h can occur on nouns referring to persons who are not small (or young). 22In ( 16) above (section 5.3), for instance, the referent is the speaker's deceased husband, a relatively tall and strong man who passed away in his eighties.In this case, the diminutive only has an emotional meaning, expressing affection tinted with compassion.As in Dalabon (section 7.1), the denotational meaning can also be absent when the base refers to a nonhuman.
As for verb complexes, the three occurrences of kanja(ng)h as a postverbal free form are clearly emotional, expressing compassion for the main participant of the event, as in (26) (repeated from ( 21)).In (26), the context rules out the 'little bit' interpretation (since the speaker is in fact complaining that they had to stay alone for a long time).Denotational interpretations of postverbal kanja(ng)h-'do X a little bit'-never surfaced in spontaneous occurrences, and were only very marginally accepted in elicitation.In ( 27), the meaning 'small' is absent; the speaker, who is over eighty years old, is talking about herself, and she can hardly be described as short, small, or young.Hence, the prevalent meaning in these postverbal occurrences is emotional, and the denotation meaning is backgrounded, if present at all.To conclude, the denotational meaning 'small' of the Rembarrnga diminutive is always optional; it can be inactive whether the item follows a nominal or a verb complex.That is, whether with nouns or with verbs, nondenotational values (emotional, pragmatic, or both) can always be the only functions of the Rembarrnga diminutive.

Summary.
As summarized in table 1, the Dalabon and Rembarrnga diminutives differ in the relative prevalence of denotational vs. emotional and pragmatic values, and in how this combines with morphosyntactic distribution.In Dalabon, denotational meanings are prevalent with human nouns, while emotional (and pragmatic) values are prevalent with verbs and nonhuman nouns.In Rembarrnga, there is no clear differentiation in this respect, as emotional values are always prevalent.The diminutives =wurd in Dalabon and (¤)kanja(ng)h in Rembarrnga both follow their hosts and occur frequently on nouns referring to children.At first sight, a large number of tokens appear in similar contexts and with comparable meanings in both languages.However, in spite of significant grammatical and semantic convergences between Dalabon and Rembarrnga, the diminutives have developed from different roots.They are effectively different linguistic tools and display morphosyntactic as well as semantic nuances.As shown in table 2 below, they differ in form, etymology, morphological status, denotational semantics, distribution, and semantic prevalence-but the emotional meanings, on the other hand, remain the same.
The Dalabon diminutive =wurd is etymologically related to nouns that mean 'child', and its denotational meaning is 'young'.The form is best described as an enclitic and occurs on most parts of speech, including verb complexes.Which type of meaning is prevalent depends on the base to which =wurd is encliticized.In the most frequent case where the base is a noun referring to a human being, denotational meanings are prevalent; on other parts of speech including verb complexes, the nondenotational meanings-emotional and pragmatic-are prevalent.In Rembarrnga, the diminutive (¤)kanja(ng)h can be a prosodically free word, but when it is bound it is better described as a suffix, rather than an enclitic.The form is etymologically derived from what has been described in the literature as an adjective meaning 'small'-and indeed the denotational meaning of the Rembarrnga diminutive is 'small', rather than 'young' as in Dalabon.
The Rembarrnga diminutive occurs mostly on nouns and its value is predominantly emotional irrespective of the morphosyntactic status of the host.

Ö
Remarkably, the nondenotational meanings of the two enclitics, and in particular their emotional values, coincide in spite of all their other differences.Perhaps because Dalabon and Rembarrnga are spoken in the same cultural context, the emotional values in particular display remarkable consistency: diminutives in both languages express compassion, affection, and endearment when witnessing intimate routines.

Distribution and expressive potential.
While the emotional meanings of the Dalabon and Rembarrnga diminutives remain the same in the sense that both these forms can express the same emotions, differences in their distribution and semantic prevalence could correlate with differences in the way speakers use these forms.Dalabon speakers can encliticize =wurd to a verb complex, expressing an emotion about an entire event.When Rembarrnga speakers produce comparable descriptions, the form kanja(ng)h is a free word.It is therefore more likely to move around the sentence and become, for instance, a free-standing interjection which could, unlike an enclitic, express emotions independent of the linguistic description of a fact.
Importantly, as a combined result of nondenotational meaning and semantic prevalence, Rembarrnga speakers can use (¤)kanja(ng)h to express affection for people who are old and not small-both because (¤)kanja(ng)h means 'small' rather than 'young', and because (¤)kanja(ng)h is primarily emotional even when the base is a noun, so that its denotational sense can be left inactive.This is illustrated in ( 16) above (section 5.3), and in (28) (repeated from (18) for convenience), where the speaker minimizes the potentially negative effect of her statement contradicting a senior relative she has respect for.This interpersonally significant usage does not occur in Dalabon.Although minimizing functions are available, the diminutive =wurd is primarily denotational when the base is a noun.Therefore, the denotational meaning 'young' always applies, and =wurd does not occur on nouns referring to older people.Thus, in spite of comparable nondenotational categories and dominant contexts of usage, the Dalabon and Rembarrnga diminutives may also occasionally carry out noticeably different discourse and interpersonal functions, due to their respective distribution.In this case, both denotational semantics and formal properties governing distribution affect what speakers can or cannot say and do with their respective expressive forms, even though the emotional semantic categories expressed by these forms are essentially identical.

Linguistic properties affecting expressive resources.
The different etymologies and resulting phonological make-up of the Dalabon and Rembarrnga diminutives may account for some of the differences between the resulting tools, and in particular between their expressive functions.Semantically, the difference in denotational meanings matches those of the etyma, with wurd 'child' extending to 'young' in Dalabon, and kanja(ng)h 'small' remaining 'small' in Rembarrnga.With respect to morphosyntactic distribution, the occurrence of Dalabon =wurd on verb complexes may have been facilitated by its shape-a single syllable starting with an approximant.This approximant sometimes merges phonologically with the final vowel or nasal consonant of the host (section 3.1) and, indeed, in Dalabon many verb complexes end with a nasal consonant.By contrast, Rembarrnga (¤)kanja(ng)h has two syllables and an initial occlusive.While this clearly did not prevent it from binding with nouns, which are rarely more than three syllables, this may be an obstacle with verb complexes, which are rarely less than four syllables. 23This may help explain why, after verb complexes, kanja(ng)h is a free form comparable to an interjection.To that extent, the particular linguistic properties of these diminutives affect their semantics, their distribution, and therefore their expressive potential.In future research, the effects of these differences in terms of expressive potential could be explored experimentally-for instance, using communication chains where successive speakers are asked to tell a story to one another in turn, and the content of the final version is compared with the initial one (Sidnell and Enfield 2012;Bebbington et al. 2017).This would shed a new and very interesting light onto the role of language in determining communication, and onto the linguistic relativity hypothesis more generally.

Conclusion.
Dalabon and Rembarrnga, both Gunwinyguan languages, share much of their grammar and semantics and are spoken in very comparable sociocultural contexts.Both languages have postposed diminutives, which, at first glance, occur in similar contexts, namely, qualifying children.However, closer analysis reveals that the two linguistic tools differ in many ways.The Dalabon diminutive enclitic =wurd is a morphologically flexible all-purpose diminutive with predominantly denotational semantics on human nouns and predominantly nondenotational semantics on all other parts of speech.The Rembarrnga diminutive (¤)kanja(ng)h, on the other hand, is primarily a nominal suffix with predominantly emotional semantics.It rarely occurs on verbs, and when it does, it is not a bound but a free form.
An important observation is that in spite of notable differences in their formal properties and denotational semantics, the emotional meanings of these two morphological tools are remarkably similar.The Dalabon diminutive and the Rembarrnga diminutive occur in a very comparable range of quite specific emotional contexts, expressing compassion, affection, and endearment when witnessing intimate routines.As shown by Ponsonnet (2014b: chapter 4) for Dalabon, and by more recent data in Rembarrnga, Bininj Gun-wok, and Kriol, the local English-based creole, this threefold combination of emotions is typical of expressive resources in this culturally unified part of the Gunwinyguan region.In this respect, emotional semantics seems relatively impervious to formal variation.In all languages in this region, expressive resources of different types (evaluative morphology, interjections, prosody) consistently express compassion, affection, and endearment when witnessing intimate routines.While these three semantic categories are attested emotional extensions of diminutives across the world (Ponsonnet 2018a), their systematic association is not particularly widespread, and may, in fact, be an areal feature.Further research will be necessary to confirm whether it is specific to the Gunwinyguan region, or applies to other language families on the Australian continent.
In spite of major resemblances in their nondenotational meanings, the formal linguistic properties of the Dalabon and Rembarrnga diminutives seem to impose differences in what speakers can express with their respective tool in each language.Differences in distributional properties constrain the contexts in which the diminutives can be used, and therefore when and with respect to whom they can express emotions.What differences such linguistic nuances imply in terms of communication potential, how this impacts interactions between speakers in each language, and whether speakers' attitudes towards emotions may be modified in consequence are questions for future research.
The above conclusions about the nature of emotional semantic categories and the impact of linguistic properties upon the expression of emotions are based on expressively rich data collected among female speakers using nonstandard documentation methods.With respect to Gunwinyguan diminutives in particular, traditional methods of language documentation had failed: previous linguistic descriptions had largely overlooked diminutives and their emotional values.This failure may be explained by an asymmetry in the gender of speakers from whom the data is collected, as well as by an (equally gendered) imbalance in the genres privileged in language documentation.These biases show that standard data collection methods in field linguistics are not particularly well tailored to capture expressive features, which typically flourish in female speech.For our understanding of expressivity across languages to progress, linguists need to update their methods in this respect.

Notes
Acknowledgments.I am immensely grateful to the language communities and consultants who helped me throughout this project, in particular Nellie Camfoo and Dorothy Cameron.Transcription.See tables A1-A3 in the appendix for the orthographic transcription used for Dalabon and Rembarrnga in this article.

Figure 3 .
Figure 3. Molly in Rabbit-Proof Fence (see n. 1).(Photo courtesy of Jabal Films Pty Ltd.) young women looks after her grandmother.'(20120719a_001_MT 205[Film]) [about the heroine of a movie providing daily care for her grandmother]

Figure 5 .
Figure 5. Ridjimaril in Ten Canoes.(Photo courtesy of Vertigo Productions.)7.4.Dalabon =wurd: summary.The semantic prevalence of the Dalabon diminutive =wurd shifts when it is encliticized on a nonhuman noun or a verb complex, as opposed to a human noun.With human nouns, the denotational meaning is prevalent; it is always present, and nondenotational values .A-DU-night-see:REDUP-PCONT 0yarra¤bba¤/nini 0kanja:ngh.1.EXCL.A-DU-sit/be:PRS:REDUP DIM 'Us two used to stay in the dark throughout the night, poor things.' (20150826_000_DC 356 [Narr]) [living alone after the rest of the family has left the station camp] -eyes-tears-flow-PPCT 'I cried about it, poor me, I shed tears about it.'(20150815_001_NC 055 [Narr]) [about an old tree that was cut and removed near the speaker's home]
Me¤kanjah badjdjan narra¤bba¤ngu¤na.Although the Dalabon and Rembarrnga diminutives occur most frequently in contexts that appear very similar at first sight, the two items have subtly different denotational meanings: the Dalabon =wurd means 'young'; . . .

Table A1 . Consonants and Their Orthographic Notation
FRICATIVEH NOTE: H is only attested for Dalabon.