A typological approach to intersubjective uses of the Finnish clitic markers =hAn and =se from the perspectives of engagement and their interrelations with subject person

2.1 The Finnish demonstrative and personal pronoun system

The modern standard Finnish tripartite demonstrative system traces back to the reconstructed Proto-Finnic system reconstructed by Larjavaara (1986, 69–75), consisting of the following demonstrative paradigm, with the n-series being the plural forms: *tämä/*nämä(t) [proximal], *too/*noo(t) and *taa/*naa(t) [distal], and *se (< *śej)/*ne(t) [medial]. As introduced in Section 1, these demonstratives can also be used as third person pronouns alongside the primary third-person pronoun *hän. A consensus is that *hän likely goes back to the Proto-Uralic anaphoric pronoun *son, etymologically corresponding to Saami *sun and Mordvin *son ‘s/he’, for instance, even though the front-vowel harmony in Finnic *hän makes the case for a Proto-Uralic reconstruction rather controversial (SKES 1987[1955], 97–8, SSA 1992, 208). Alternatively, we can also consider that the reason for the front-vowel harmony in Finnic *hän is the result of analogy from the predominantly front-vowel pronoun paradigm, considering the symmetrical -e rhyme in the plural forms.

Regarding the semantics of the Finnish demonstratives, the classification may vary according to the research tradition. For instance, a spatiality-based approach (e.g. Diessel 2013) divides Finnish demonstratives into two distance categories: tämä as near to the speaker, while se and tuo as far from the speaker. Nevertheless, spatiality alone cannot explain the contrast between the non-proximal demonstratives se and tuo, because their difference is not exclusively based on the distance from origo, i.e. the speaker, but can also be alternatively classified based on logophoricity, i.e. how a referred entity is related to the speaker. Another interactional approach, meanwhile, rather pays attention to a discourse-sphere orientation: tämä for the speaker’s sphere, tuo for the remote sphere, and se for the hearer’s sphere (e.g. Laury 1997, 59–60). Lying between these two approaches, Larjavaara (1990, 95–100) illustrates that the three Finnish demonstratives can be classified on the basis of both spatiality and logophoricity: tämä as speaker-approximate, tuo as speaker-approximate and speaker-centred, and se as hearer-centred. In the current study oriented towards the emerging trend of engagement, our interpretation which we further use for analysing the clitic use of se is that se typically refers to a referent which is not in the speaker’s sphere, but the speaker uses it to draw the hearer’s attention and establish a joint attention (see also Evans et al. 2018b for the engagement-based approach to demonstratives).

When used as third-person pronouns, tämä and tuo, respectively, are referential devices referring to entities which are newly mentioned and currently being discussed. At the same time, se as well as hän have been analysed as referential devices serving an anaphoric function for an entity which is identifiable and mutually understood between the interlocutors (e.g. Etelämäki 2006, 14–5, 2009, Priiki 2017). Furthermore, previous studies have discussed the difference between hän and se, which relate to logophoricity (e.g. Saukkonen 1967, Laitinen 2002, 2005, Hakulinen et al. 2004, §1469, Priiki 2017). As illustrated in (2a) vs (2b), hän serves a logophoric function referring to a main clause activity performer, while se can also exophorically refer to another person.

In (2a) and (2b), the logophoric hän refers to the main activity performer, Juhani, the opposite of which, Marja as the message recipient, is referred to with se (see also Ikola 1960, 19–63).

In any case, the functions of the two pronouns may vary across discourse situations and can sometimes overlap across contexts of use (Hakulinen et al. 2004, §1470). In narratives, for instance, the speaker uses hän to switch their role as narrator to the participant of the discourse and to express attitudinal functions when referring to a person with politeness, negligence, wonder, or suspicion (Vilppula 1989, 398–9). What is relevant for the analysis of their clitic uses is that hän refers to entities in a higher degree of the interlocutors’ attention with a broader range of epistemic status and speech acts, while se has a stronger referential force for capturing the hearer’s attention and managing contrast among identifiable entities.

2.2 Formal and functional comparison between =hAn and =se

The clitic uses of hän and se in the function of expressing interlocutors’ (inter)subjectivity show several differences when it comes to the questions of form and function. In the actual language use as described in the Finnish reference grammar (Hakulinen et al. 2004, §827, §830–2) and as observed in the Suomi24 corpus, formal differences between =hAn and =se can be generalised in the aspects described below. The generalisation in this section is supported by quantitative data from corpus search where relevant (see the full quantitative data later in Section 4.2).

In terms of morphology, =se may inflect for number in its plural form =ne from ne ‘those; they’, as being contrasted in Examples (3) and (4), while =hAn does not inflect in the plural form **=he, which would be the normal plural form of the pronoun hän:

Plural form =ne, however, occurs mainly with common nouns, as the plural form =ne does not co-occur with plural pronouns alone without =hAn although the frequencies are extremely low (see Appendix 1). In relation to subject persons, =se is sensitive to the person in the plural as the forms like me=se [2sg=se] (à 38/101,979 occurrences) or te=se [2pl=se] (à 61/109,972 occurrences) are rare but not impossible, as shown in Examples (5) and (6). Meanwhile, the occurrence of =hAn is not restricted by person in any way.

Regarding the host attachment, =hAn can co-occur with any part of speech, including adverbials (7) and verbs (8), while there are clear restrictions on the use of =se, such as after numeral **kolme=se [three=se].

As for constituent order, =hAn always precedes =se if both simultaneously occur on a word, yielding a conjoint form =hAn=se as in Example (9) (see also Karttunen 1974 for the organisation of multiple clitics in Finnish).

In orthography, =hAn is always written as a part of the host word minahän [1sg.han], which is not the case for =se almost exclusively written as a separate word. However, conjoint forms are also observed in the corpus: minäse [1sg.se] (à 9/101,979 occurrences), sinäse [2sg.se] (à 27/109,972 occurrences), and hänse [3sg.se] (à 31/120,386 occurrences), albeit the annotators of the Suomi24 corpus, interestingly, leave the morphological analysis of these forms undefined as ‘other unknown’ morpheme [OTHER_UNK].

Based on the observations above, it seems that =hAn has been grammaticalised and normativised as a genuine clitic to a larger extent (as recognised in Hakulinen et al. 2004, §830). The higher degree of grammaticalisation is manifested also in the fact that the form and function of =hAn have been studied more extensively (e.g. Liefländer-Koistinen 1989, Duvallon and Peltola 2012, 2013, Duvallon 2014), while there are practically no studies dedicated to the formal and functional description of =se. From the usage-based perspective, the frequency of use regarding the two clitics in the corpus also confirms this view, as =hAn appears in 85.83% of co-occurrence with personal pronouns, but only 11.06% for =se, and 3.11% for the conjoint form =hAn=se (see Section 4.2). Related to the last point in the observation above, the different degrees of grammaticalisation between the two clitics are also reflected in how Finnish language users spell out these in written forms, given that =se written as part of a word as sinäse [2sg.se] is very rare, as it is usually written separately from its host word as sinä se [2sg se].

In terms of functions, previous studies consider =hAn as the speaker’s stance-taking device which the speaker uses for drawing the hearer’s attention to a specific state of affairs (e.g. Halonen 1996, Hakulinen et al. 2004, §830–2, Duvallon and Peltola 2012, 2013, Duvallon 2014). This at least to some extent resembles the description of the Swedish ju (Bergqvist 2020). Typical context of the use of =hAn concerns statements about topics which are related to knowledge, view, or assumption on the given state of affairs, often being shared between the interlocutors and for which the speaker expects some feedback or following actions from the hearer. This concerns speech act functions, such as disagreement (10), seeking confirmation, reminder, or warning (11).

In contrast, there is previously no study specifically on functions of =se, but functions of its partitive form sitä have been discussed to a certain extent (Hakulinen 1975; Hakulinen et al. 2004, §827). As the use of sitä described in previous studies resembles the results obtained from the investigation of =se in the current study (Sections 4 and 5), we see a good reason to provide the following description which is applicable for both forms. Namely, they construct a claim based on concrete observable evidence, referring to a state of affairs which the speaker expected to occur (12a), against counter-expectation marked by =hAn in (12b), or bringing an inclusively or exclusively observed state of affairs to inclusive attention with the hearer (13). The Finnish linguistics literature also suggests several other affective meanings for sitä, which can express a playful or disapproval attitude.

The marker =se (and its partitive form sitä), accordingly, behaves as a marker for confirming novel information or pointing out contrast against other alternatives, as is expected from its properties inherited from the demonstrative origin (Himmelmann 1996 discussed in Section 2.3).

The description of the two clitics will establish a baseline for exploring other potential functions of =hAn and =se using the data in Section 4, which will provide concrete evidence of functional resemblances between =se, described in this study, and sitä as discussed in the Finnish linguistics research literature. Before delving into the data analysis, Section 3 will discuss and establish the framework of engagement as a tool for subsequent analysis.

3.1 Introduction of engagement as a functional category

The concept of engagement was introduced to describe a linguistic phenomenon of intersubjective and epistemic management involving the interlocutors’ knowledge and attention in the discourse, initially under the French term assertif by Landaburu (1979, 111–26, 2007) for Andoke, an Amazonian language spoken in Colombia. Andoke has four different verb prefixes for signalling contrastive configurations of the speaker’s (S) and the hearer’s (H) epistemic perspectives in terms of certainty towards a state of affairs (+ yes vs − no): 1) kē- = catégorique [S+H−], 2) b- = positif [S+H+], 3) k-/d- = non-savoir [S−H+], and 4) ma- = probable [S−H−] (see a further typological classification in Section 3.2).

The phenomenon is also observed in other Amazonian languages, such as Nambikwara and Kogi, as shown in Examples (14) and (15).

In Nambikwara, an unmarked declarative utterance in (14a) may assume the speaker’s authority on a state of affairs, while the use of a morpheme -ti ² .tu ³ - in (14b) marks the information as shared between both the speaker and the hearer. In (15a), in turn, the speaker expects the hearer to know x while the speaker is unaware of x, indicating hearer authority. Meanwhile, in (15b), the speaker expects the hearer to know x, and the speaker knows x too, pointing towards the interlocutors’ shared knowledge. The epistemic (a)symmetry between the speaker and hearer is thus encoded in Kogi morphologically with two distinct morphemes: sha- [asymmetrical] vs shi- [symmetrical]. Examples (14) and (15) bring into attention the two aspects of interlocutors’ intersubjectivity: 1) epistemic authority and 2) epistemic (a)symmetry (see the most recent view on these epistemic contexts in Section 3.2).

Observations of different strategies to code the distribution of knowledge between interlocutors in individual languages lead to a cross-linguistic studies and typological conceptualisation of engagement by Evans et al. (2018b).

Engagement refers to a grammatical system for encoding the relative accessibility of an entity or state of affairs to the speaker and addressee. (Evans et al. 2018b, 142)

In its intersubjective sense, engagement targets the epistemic perspectives of the speech act participants, signalling differences in the distribution of knowledge and attention between the speaker and the hearer (see also a similar concept of ‘territory of information’ in the study by Kamio 1997). As such, it specifies whether information is shared or exclusive to one of the speech act participants (Bergqvist and Kittilä 2020b), as illustrated in Nambikwara examples (14a) vs (14b).

Asymmetry evolving in the interlocutors’ states of knowledge and attention can also be utilised as a means for the speaker to take or give epistemic authority to the hearer (Evans et al. 2018a, 118), as is seen in Kogi Examples (15a) vs (15b). This phenomenon has some resemblances to the degree and hierarchy of accessibility and givenness observed in the use of (in)definite articles and possessive pronouns (see a synthesis shown in Abbott 2004, 122–4). From the perspective of intersubjectivity, (in)definite articles and possessive pronouns can express a neutral state (16a) or (a)symmetry of knowledge and attention between the interlocutors (16b, 16c).

The use of indefinite article in (16a) is inquired by the speaker that the mentioned entity is not in the hearer’s attention yet, while the use of definite article in (16b) presumes that the hearer is already aware of the entity being specific among other possible identifiable alternatives through presupposition from the earlier discourse context. The use of possessive pronoun in (16c), on the one hand, entails that the hearer has the mentioned entity in awareness and memory, but the speaker, on the other hand, ensures that the hearer can recognise this specific entity by using an explicit expression through possession.

3.2 Towards a typology of epistemic perspectives and the interlocutors’ engagement

The attempt to typologise the engagement system based on epistemic perspectives between the interlocutors was already present as a quadripartite typology in the Andoke grammar by Landaburu (1979, 119, 2007) and has been recently put forward towards an operationalisation by Grzech (2022). The classification is based on how epistemic authority (+ yes vs − no) is distributed among the speaker (S) and the hearer (H), as mapped in Figure 1. Note that the original version by Grzech (2022) has A(ddressee) for the message receiver, instead of H(earer) used in the present study for the same purpose.

Figure 1

Knowledge distribution in interaction (adapted from the study by Grzech 2022).

Type 1 is labelled as ‘uninformative’ (positif in Landaburu 1979) which relatively well captures the informativeness status of the case, that is, an expressed proposition is not novel and provides no new information for either of the speech act participants [S+H+]. However, the rather uninformative nature of these statements does not mean that they are irrelevant to the discourse and would not be attested in normal language use. An example is provided, for example, by an evaluation scene where the two interlocutors have the same information, and they compare pieces of information with each other. No new information is provided, but the communication still has a clear goal.

Type 2, ‘assertion’ (catégorique in Landaburu 1979), is perhaps the most expected pattern of natural communication since the speaker holds the epistemic authority and informs the hearer of a novel piece of information [S+H−].

In Type 3, ‘interrogative’ (non-savoir in Landaburu 1979), the speaker is searching for information themselves and believes the hearer to have the required/necessary information [S−H+].

As for Type 4, ‘evidentiality’ (probable in Landaburu 1979), neither the speaker nor the hearer has epistemic authority [S−H−], which means that no real exchange of information takes place (neither provides, nor receives any novel information). However, the evidentiality type of context is also rather common and is attested in cases where the interlocutors speculate about possible outcomes of events. For example, before a (live) sport event, neither the speaker nor the hearer can actually know who is going to win the competition, but for the sake of interpersonal communication, it may nevertheless be interesting to speculate about the result. This type also shows that communication is not only for conveying (novel) information from the speaker to the hearer, but it has numerous other conversational functions.

In any case, not all languages have dedicated grammatical markers for all types of epistemic perspectives like Andoke (as described in Landaburu 1979, 2007), and some of them can be inferred from contexts. For instance, English speakers use tag questions for expressing Types 1 and 4, while Types 2 and 3 are interpreted from simple assertive and interrogative statements. Upper Napo Kichwa, in turn, has dedicated verb endings to express each of the types above, as shown in Example (17) where (17a) is a baseline context while the subsequent examples give extended epistemic contexts which may correspond to some of the types discussed above.

As is shown in (17), the proposed framework based on epistemic authority and (a)symmetry allows for a more specific description of epistemic contexts across languages and for tracing the pragmatic extension of grammatical elements towards engagement.

3.3 A cross-linguistic view on pragmatic extension towards engagement

Observations on grammatical marking of interlocutors’ knowledge and attention in individual languages in an engagement-like fashion were made previously in connection to various domains of language structure and discourse, such as time (tense), attention (accessibility), knowledge (epistemic authority), and identifiability (determiner) (as given by Evans et al. 2018b, and see Grzech 2022 for epistemic authority). Given the intersubjective nature of engagement, the pragmatic extension towards engagement uses often involves constructions with etymologically referential lexemes, particularly deixis (Janssen 2002, Diessel 2006, Kratochvíl 2011) and person (Dahl 2000, Bergqvist and Knuchel 2017, Schultze-Berndt 2017, Knuchel 2019, Bergqvist 2020).

Such a tendency of referential elements becoming markers of engagement likely roots in deictic and anaphoric characteristics of these semantic categories, which have developed discourse-pragmatic uses as cognitively determining the positional engagement of speech act participants from the state of affairs involved. In other words, something that is deictically closer to the speaker can also be viewed as being cognitively closer to the speaker’s knowledge, attention, and stance. This has a close relation to what Himmelmann (1996) labels as a ‘recognitional use’ of demonstratives in the following principle.

Recognitional use … involves reference to entities assumed by the speaker to be established in the universe of discourse and serves to signal the hearer that the speaker is referring to specific, but presumably shared, knowledge. It invites the hearer to signal the need for further clarification regarding the intended referent or to acknowledge that he or she, in fact, knows what the speaker is talking about. (Himmelmann 1996, 240)

This is one clear instance of inherited semantics, which undergoes an extension towards more advanced pragmatic uses in the universe of discourse. Such a mechanism of pragmatic extension is the case for the Finnish clitics =hAn and =se investigated in the current study. In particular, hän as an anaphoric pronoun pointing to neither the speaker’s nor the hearer’s spheres makes it accessible to both speech act participants, corresponding to the recognitional characteristic described above.

Previous studies on the engagement-like uses of various etymologically referential words also report similar findings beyond the South American languages discussed in Sections 3.1 and 3.2, for instance, for Swedish particles ju and väl (Aijmer 1977, 1996, Eriksson 1988, Teleman et al. 1999); Spanish independent si conditional clause (Schwenter 1996); German modal particles ja, wohl, doch, and etwa (Waltereit 2001, Gast 2008); Abui free-standing demonstratives (Kratochvíl 2011); Vietnamese demonstrative-derived sentence-final particles (Lê 2002, Adachi 2016); and Finnic and North Russian postposed demonstratives, respectively, -se and -to (Yurayong 2020), among others. These studies have shown that the mentioned markers of engagement participate in the epistemic management of the availability and exclusivity of interlocutors’ knowledge and attention through inclusive and exclusive experiences.

Note that in some other languages, interlocutors’ engagement in the discourse, involving attitude and control of epistemic perspectives, can also be coded in verbal conjugation, distinguished between the interlocutors’ opinion (egophoric) and a matter of fact (allophoric) (Bergqvist and Knuchel 2017, 369). This has been initially reported as the case for Kathmandu Newar (Hargreaves 1991), and later scholars also make similar remarks in languages of their expertise, particularly South American and Sino-Tibetan languages (see the collective volume on egophoricity in Floyd et al. 2018). However, such grammatical strategies will not be discussed in the present study which exclusively focuses on grammatical elements derived from referential lexemes.

To discuss several significant case studies, investigation on the co-occurrences of intersubjective markers, derived from referential words and serving the function of expressing interlocutors’ engagement, has been previously conducted by Bergqvist (2020) for Swedish modal particles ju and väl, the basic functions of which are illustrated in Examples (18) to (20). Note that it is ju which is derived from an indexical element: Swedish ju ∼ Danish jo ← Middle Low German jo < Proto-Germanic *ja ‘thus, so’, while väl is a cognate to English well, sharing the same root as will, in the sense of ‘something desirable’.

The results (Bergqvist 2020, 490) show that ju as a marker for epistemic asymmetry [S>H+] in (18) or lack of attention, i.e. mirative use [S−H−] in (19), co-occurs with subject persons in the following order from most to least frequently: third person (49%) > first person (37.5%) > second person (13.5%). Regarding väl in (20) as a marker for epistemic symmetry [S+H+], the order from most to least frequent co-occurrence is as follows: second person (39%) > third person (32%) > first person (28%). In this respect, there is a similar frequency trend between the Swedish väl and Finnish =se, which might be due to their functional similarity as markers of hearer-oriented shared stance, yielding epistemic symmetry between the interlocutors. At the same time, the frequencies point to the fact that the Swedish ju is more common with second and third persons than with first person (see similar results for Finnish in Section 4).

Another interesting parallel is Vietnamese sentence-final particles which derive from demonstratives. The spatial dimensions of demonstratives namely extend their indexical and referential uses to the discourse-pragmatic domain for expressing interlocutors’ engagement, the pragmatic extension previously discussed by Adachi (2016), as summarised in Table 1.

Differences in sentence-final uses of the demonstratives are given in Examples (21) to (24).

The proximal đây in (21) recurs the statement to the employee’s attitude towards the difficulty of the assignment evolving from her direct experience [S+H−], whereas the medial đấy in (22) expresses the mother’s inference on the interest and relevance of the statement to her daughter [S<H+]. In (23), the other adnominal medial ý emphasises the sharedness of the knowledge between the two interlocutors [S+H+]. As for (24), the distal cơ expresses epistemic asymmetry between the speaker and hearer, with the speaker holding a higher degree of knowledge over what could be the hearer’s counter-expectation [S>H+], i.e. mirative use (DeLancey 1997).

Comparing the Vietnamese sentence-final demonstratives to the two Swedish particles, Vietnamese proximal đây and Swedish ju have some functional properties in common. For these speaker-oriented markers, all of them typically express the epistemic asymmetry between interlocutors often entailing the involvement of the first person as a speech act participant who takes their own stance [S+H−/S>H+]. In presenting the known fact about the third person to the hearer, however, the Swedish ju may also express the interlocutors’ symmetry of not having the authority and knowledge on a state of affairs in the mirative uses [S−H−], which would correspond to one of the functions of the Vietnamese distal kia/cơ (see remarks below).

In contrast, the Vietnamese medial adnominal ấy/ý is more comparable to the Swedish väl in the sense that they base the uttered statement on interlocutors’ shared knowledge yielding epistemic symmetry [S+H+], with a stronger orientation towards the hearer sphere. They more often appear in statements related to the second > third person, crucially not the speaker (i.e. first person), or to their joint knowledge and attention. For Vietnamese, this might result from the semantics of its lexical form as medial demonstrative, which can also be alternatively analysed as hearer-centred demonstratives, thereby signalling the involvement of the second person in the epistemic sphere of the discourse.

As for the Vietnamese medial đấy (independent), it differs from the medial adnominal ấy/ý and Swedish väl in that its intersubjective status is epistemic asymmetry with the hearer holding a stronger degree of epistemic authority and knowledge [S<H+]. Meanwhile, the distal kia/cơ differs from the proximal đây and Swedish ju in that it can point towards either symmetry or asymmetry and with no person orientation due to its original deictic distality, while the epistemicity degree can be absent in the case of mirative uses [S−H−], or higher towards the speaker when attempting at establishing joint attention [S>H+].

The comparison between Swedish and Vietnamese reveals that languages with a larger set of intersubjective markers, such as Vietnamese, can distribute dedicated engagement functions more specifically across various grammatical resources. In contrast, languages with less extensive intersubjectivity marking, like Swedish, may employ a single marker for multiple epistemic contexts. In such cases, the contextual interpretation, including interrogatives, becomes a factor when determining its function.

Next, we go beyond the epistemic (a)symmetry issues to explore in the Finnish data whether subject persons play any role similar to the description and specific Swedish and Vietnamese cases as discussed in this section.

4.1 Appropriateness judgement by language users

Based on the observation of formal and functional differences presented in Section 3, we further examine the validity of interrelation between the clitics =hAn and =se, and other functional categories. The first part of the study is conducted through appropriateness judgement among 35 bachelor’s and master’s students at the Department of Languages, Faculty of Arts, University of Helsinki, who are L1 and L2 speakers of Finnish. The language of instruction in the course where the test was conducted is Finnish, which means that the level of command of Finnish of all participants is very high, although there were also a couple of non-native Finnish speakers in the teaching group.

The participants were asked to choose which of the two markers, =hAn or =se, fit better in given contexts of use. They could also voluntarily provide more specific explanations as to why one marker was more suitable in the context than the other. The questionnaire includes 40 questions, featuring very different kinds of scenarios with very different kinds of evidence. We chose both contexts which, in our own view, clearly favour =hAn and contexts, where, again based on our subjective evaluation, =se is clearly more appropriate. In addition, we also created scenarios, where both seem more or less equally likely and scenarios between all these types. The scenarios varied, for example, based on whether the evidence is concretely present or not and whether the speaker expects the hearer to share the given information or not. Moreover, the grammatical person varied between the scenarios.

Each question is given in the same format, starting with a given context and then the reaction in which the clause-second slot is left blank for the selection between =hAn and =se. This should make the genre of the questionnaire comparable to the corpus part of the study, based on the online social networking website Suomi24 used in Section 4.2, as the internet discussion is similarly structured in such a pattern in which the first post stimulates reactions from the audience.

Figure 2

Suomi24 discussion interface (https://keskustelu.suomi24.fi/t/17595758/olen-sekoillut-suomi24ssa).

The results show crowdsourced tendencies whether =hAn or =se is preferred in the given contexts. The answers from 35 participants are organised in a three-value-based form: =hAn = 0, both = 1, and =se = 2. The results are subsequently accumulated, according to which the use of =hAn is preferred among the participants when the average is closer to 0, and vice versa closer to 2 for =se. In this section, we discuss the occurrence of =hAn and =se in connection to different subject persons. We primarily provide quantitative results based on the frequency of uses with a brief summary of the participants’ comments on the appropriateness of the markers in given contexts, while the qualitative analysis of interrelation between markers of engagement and subject persons will follow in Sections 5 and 6.

4.2 Usage-based account on distribution among different subject persons in the use of =hAn and =se

As a second part of our study, the observations from crowdsourcing in Section 4.1 are tested against the actual language data retrieved from the Suomi24 Sentences Corpus 2001–2020 (Suomi24), using the Language Bank of Finland’s web interface, Korp, for searching. The corpus consists of 20 sub-corpora from each year’s version of the website Suomi24 from 2001 to 2020, including a total of 4,582,558,555 tokens and 391,965,356 sentences.

As mentioned in Section 4.1, the corpus is comparable to the contextual setting in the questionnaire part, as both involve situations in which a statement is given as a foregrounding context in which the audience reacts, as illustrated in Figure 2. Furthermore, Suomi24 is an open and anonymous platform, which often results in heated discussions. The data are thus biased towards certain genres, which may skew the results somewhat. However, despite these potential problems, we maintain that Suomi24 is a suitable corpus for the goals of the present article.

As for the search queries, the following filtering solutions are applied. The sample takes only contexts where first, second, and third person pronouns co-occur with =hAn and =se, or both in the respective sequence =hAn=se, and finite verbs agreeing in person and both numbers (singular and plural) with the pronouns. Note that the queries can capture formal variations of personal pronouns, e.g. minä, miä, mä, and mää for the first-person singular pronoun. This combination of morphemes gives a total of 28 search scenarios (the full list and CQP queries are shown in Appendix 1). The set of CQP queries employed in the data search aims at maximising recalls to achieve an overall picture of frequency and, therefore, may partially sacrifice the precision of the data received. For instance, a borderline case overlapping with a cleft construction and a fronted subject pronoun, such as hän=se on, joka … [3sg=se be.3sg rel] ‘it is him/her who […]’ (as described in the study by Hakulinen et al. 2004, §1383, §1388), may have been included in the recall to a small extent. In any case, the aim is more at giving rough numbers of occurrences for the sake of comparison and confirmation with the qualitative account in Section 4.1, considering also the note in Section 1 that =hAn and =se are both completely optional elements.

The results on the distribution of the clitics and their combination =hAn, =hAn=se, and =se across different subject persons are given in Table 3.

In general, the frequency of =hAn clearly outnumbers that of =se, which supports the implication about their degrees of grammaticalisation and naturalness (as discussed in Section 3.1). This stems from the idea that elements which can be used in more contexts have been grammaticalised to a larger extent than elements which can be used more restrictively in specific contexts. The results from the usage-based data show significant correlations with the results from the questionnaire data in that the use of =hAn is more frequent than =se overall, and in terms of person, =hAn co-occurs most often with the first person who usually holds the epistemic authority towards a given state of affairs. At the same time, =se occurs more frequently with the second person, which may be due to a difference in context settings, as the online social networking website often contains a discourse context where the statement about the first person in the opening turn is reacted to by repliers as the second person with expression confirming the shared knowledge (see the structure of Suomi24 discussion platform in Figure 2). Further discussion on the rationale behind the correlations with subject persons will be given in Section 6.

At this point, we can summarise the order of frequencies for the corpus data, as follows: =hAn, first person > third person > second person; and =se, second person > third person > first person. This result will be used as the basis for discussion in Section 5.

5.1 Nature of the data and methods

The first two points of discussion concern the overall frequencies of use and the difference in contextual settings of the two methods employed in the current study. First, we should note that despite the very different nature and foci of the two methods employed, the correlations between the markers of engagement and subject persons show similar, yet not identical, tendencies. With all subject persons, =hAn is generally more common due to its significantly higher overall occurrence frequencies (as argued in Sections 3.1 and 4.1.5).

Second, the contexts have their effect observable in the use of =se, which is in general less common in the corpus-based study than in the questionnaire-based study. Again, this is probably not a coincidence, but these differences have good reasons. The most evident of these is found in the different nature and goals of the two studies focusing on functions and frequencies, respectively. For the questionnaire, we created scenarios in which either of the clitics, based on our own intuition, would be more normal, and for this, we also needed to include scenarios, where we could expect =se to be more common. This very naturally makes the occurrence of =se higher than in the normal language use observed in the corpus data which is more telling in this regard, and our findings lend further support to the claim that =hAn has been grammaticalised to a larger extent than =se (as proposed in Section 3.1). Furthermore, the fact that =se is not compatible with certain word classes may be responsible for the frequency differences here. It is naturally also possible that there are significantly fewer contexts where =se would be preferred in normal language use, but we leave this aspect for further studies in detail.

Another issue with the contextual settings is that =se is the most common with the third person in the questionnaire survey, while through the quantitative corpus-based method, =se occurs most often with the second person. As noted in Section 4.2, one of the reasons for the highest frequencies for co-occurrences of =se and the second person in the corpus may be found in the contextual structure of the online social networking platform in which many cases are such that someone states something, and someone else reacts to this. In other words, what is said first serves as a kind of concrete evidence, and the response is thus based on what was just said, and the nature of the evidence supports the use of =se as a reaction to the previous statement by the second person (Figure 2). Despite the frequency differences with regard to co-occurrences between markers of engagement and different subject persons observed in the two methods, the corpus serving as a primary representative data in the current study interestingly shows similar results with those of the corpus study of the Swedish modal particle pairs ju and väl by Bergqvist (2020, 486) discussed in Section 3.3.

As remarks on the data and methods, a more extensive cross-linguistic comparison, based on actual language use data as is done in the spirit of engagement research for Vietnamese (Adachi 2016), Swedish (Bergqvist 2020), and Finnish in the present study, can better shed light on distributional tendencies and an ultimate limit of engagement system in a language, and at a more general level improve the description and typologisation of engagement and linguistic intersubjectivity.

5.2 The role of person in the intersubjective uses of the clitics =hAn and =se

In contexts involving the first person, =hAn is more common and =se less common. There are at least three reasons for this (in addition to the general higher frequency of =hAn mentioned in Section 5.1). First, events where the speaker is basing their claim on actual concrete evidence may be rather few in number, which makes =se functionally less appropriate, as there is naturally less need for communicating something both the speaker and hearer can witness, i.e. uninformative statements (as defined in Sections 2.2 and 3.2). Second, the speaker is by default the epistemic authority (S+) with first-person statements, and this does not need to be stated explicitly, which makes it possible for the speaker to utilise engagement characteristics of =hAn to express the speaker-oriented epistemic asymmetry (as discussed in Section 5). Third, the lower frequency of =se with the first person may follow from the simple fact that person does not determine the use of the clitics to any large extent, but other epistemic-related factors discussed above are more relevant to this (see below).

As for the second person, despite the general higher tendency of =hAn, an explanation for the uses of both clitics is not as clear-cut as in the case of the first person. The main difference to the first person is that the involvement of the second person often entails the hearer authority (H+). However, =se, which favours epistemic symmetry between the interlocutors, more frequently occurs here than with the first person. This could be seen as somewhat unexpected, but there are also good reasons why =hAn is not as clearly the preferred choice as with the first person. First, =hAn is common in cases where the speaker expects the hearer to have epistemic authority, but the speaker, at the same time, also has a hunch about what they are talking about. For example, =hAn is common when the speaker tries to stimulate interaction with the hearer and expect some feedback, looking for a confirmation for what they believe to be the case, as is the case in (28). Second, it is more common that the speaker somehow reacts to what someone has stated earlier, and the previous utterance involving an interlocutor who will become the second person in the reaction turn can be viewed as a kind of concrete evidence for the one performing reaction, which would explain why =hAn does not dominate as clearly as with the first person (see the points on contextual settings made in Section 4.1).

In the questionnaire study, =se is somewhat more frequent with the third person than =hAn. The main difference between the first and second persons is that with the third-person involvement, neither the speaker nor the hearer is by default the epistemic authority, but this varies more based on who has better access to knowledge (see the comparison with non-person-oriented engagement of distal demonstratives in Section 3.3). This may be highly relevant to the more frequent occurrence of =se. First, as the interlocutors are not dealing with evidence that would inherently be more accessible to either the speaker or the hearer, the speaker tends to choose =se rather than the recognitional =hAn. Closely related to this, the interlocutors are more dependent on concrete evidence typically expressed by =se, because one cannot use their own general knowledge of the world as a basis for their claim. Second, as the epistemic authority does not inherently belong to either of the speech act participants, the speaker has more freedom to choose which of the two clitics they will use, and they may opt more often for =se for stimulating interaction, establishing joint attention with the hearer in the described situation. It is interesting to note also that the overall numbers of the clitics are highest for the third persons (as shown in Table 3 in Section 4.2), although the common nouns were not included in the quantitative study. The differences are statistically not very significant, which again affirms that person does not determine the general use of the clitics in any stringent way. However, we may perhaps note also here that with the third-person epistemic authority is not set automatically by the asymmetry of knowledge distribution between speech act participants. Consequently, we may claim that it is slightly more relevant to mark the access to knowledge, epistemic authority, and intersubjectivity with the third person, even though the differences are not very significant.

Impersonal constructions have been surveyed only through questionnaire, but we consider it worth mentioning our findings here. Namely, the very clear dominance of =hAn with impersonal constructions can be viewed as rather unexpected if we consider the dominance of =se with third person. In both cases, the third person and impersonal, epistemic authority does not inherently belong to either of the speech act participants. However, the occurrence of =hAn is clearly more common in the questionnaire than in the third person. We may speculate that with impersonal constructions, especially with those expressions describing weather conditions, humans cannot have any control over what happens, which completely excludes (personal) epistemic authority from the speech act participants and thus makes =hAn more common. In other words, despite not having the knowledge, the interlocutors are rightful to have their own opinions about what has happened, and the best they can do is to compare opinions with each other, yielding a symmetric (S−H−) situation in the evidentiality type of context where =hAn is clearly suitable (as stated in Section 3.2). In any case, it might be that since epistemic authority does not inherently belong to any of the speech act participants, other factors possibly play a more important role and instead of person we would need to look for complementing explanations elsewhere. For example, the use of certain modal verbs (e.g. pitää ‘have to’ as in (25)), the use of the imperative form, affective lexical elements (e.g. takahikiä), temporal particles, the use of exclamation mark (all as present in (26)), the use of discourse particles (nyt, sit(ten) and kyllä, as in (27)) may affect the choice of the appropriate particle. However, in this article, the focus was solely on the effects of the subject person.

To account for the hypothesis of the current study about the interrelation between markers of engagement and subject persons, the data analysis has shown that contexts accommodating different subject persons can predict the preferences towards either of the clitics, though rather partially. As the differences in frequencies of co-occurrences with different subject persons do not indicate significantly big gaps, we conclude that person is not a triggering factor, but at its best a predictive element in an utterance which defines the epistemic contexts and perspectives. Further investigation to identify controlling effects with other grammatical categories, such as tense-aspect-mood and evidence types, can be tested in future studies.

5.3 The Finnish clitics =hAn and =se in the engagement framework

By analysing functions and frequencies of use for the Finnish clitics in the typological framework based on epistemic authority and (a)symmetry (Landaburu 1979 and Grzech 2022, as discussed in Section 2.2), there are clear tendencies for how the markers are used in which case. The same model given in Figure 1 in Section 3.2 is used as a base for illustrating the functional mapping of the Finnish clitics =hAn and =se in Figure 3. The mapping is organised according to the degree of contextual suitability of each clitic and zero marking, based on observations in Section 4. As constraints caused by epistemic asymmetry, the classification in Figure 3 hints that =hAn is not suitable for an assertion scenario where knowledge is not shared with the hearer [S>H], while =se is not suitable for an interrogative scenario where the speaker seeks knowledge from the hearer (S<H).

Figure 3

Knowledge distribution in intersubjective uses of the Finnish clitics =hAn and =se.

In Type 1, uninformative [S+H+], both clitics are suitable for such epistemic symmetry, but the selection may depend on an intended speech act. As described in Sections 2.2 and 4.1, the use of =hAn is recognitional, relying on the interlocutors’ shared knowledge and experience. This motivates such speech acts as a reminder in (25) and a warning in (31). Meanwhile, the use of =se in such positive epistemic symmetry can convey affectedness, such as in irritation and disappointment in (16) or humiliation in (27) and (30). The effects of person are not overwhelmingly significant here, because the distribution of knowledge is symmetric in that both the speaker and the hearer have (equal) access to the information in question.

Type 2, assertion [S+H−], can perhaps be viewed as the prototype of normal communication in which zero marking of engagement should be more common, while the use of both =se and =hAn seems a bit odd in this case. In any case, the use of =se can be more natural and is possible in case the speaker is informing the hearer of something just witnessed and intends to pass on that novel knowledge to the hearer without expecting feedback. A representative example of this subtype can be (1b) in which the speaker declaratively informs the hearer that Kalle went fishing with President Kekkonen without considering the hearer’s prior knowledge. In this type, second-person subjects are slightly less common, because the subject referent should have the epistemic authority.

Similarly, in Type 3, interrogative [S−H+], both clitics seem less natural. The use of =hAn is, however, marginally possible for Type 3, but it usually implies that the speaker has some prior knowledge about what they are referring to, and the speaker’s epistemic authority is not absent but would rather be scalarly weaker than the hearer [S<H+] as the speaker seeks the hearer’s feedback or confirmation on the state of affairs under question. It seems that the clitics are not very felicitous with normal statements (Type 2) and questions (Type 3), but they occur whenever both the speaker and the hearer have some knowledge about what is talked about, but the knowledge is not equally distributed. An example for this context is reminder and warning in (11) in which both the speaker and the hearer know about the hearer’s allergy to fish, but the hearer’s epistemic authority should be stronger. This makes the second person a more felicitous subject than, for example, in Type 1.

As for Type 4, evidentiality [S−H−], both clitics appear in such a negative epistemic symmetry and often convey mirativity, observing that something exceeds the expectation. The selection is based on how evidence is acquired and whether the acquisition is exclusive or inclusive. On the one hand, =hAn more often appears in contexts where the knowledge is mutually shared between the interlocutors, acquisition is inclusive, and the speaker tries to stimulate interaction as in (12b). On the other hand, the use of =se is not bound to previously shared knowledge but the speaker informatively shares a personal direct observation with the hearer without requiring feedback as is the case in (12a) in which they both see that it is raining at the moment of utterance. Type 4 can also be called a ‘speculative’ type since without any concrete information the best the interlocutors can do is to speculate. The use of =hAn in this case would mean that the speaker is making guesses which they expect the hearer to share the stance of. Both first- and second-person subjects are rather infelicitous in this type because neither the speaker nor the hearer has epistemic authority.

What we have observed in the Finnish data differs from the Swedish corpus data and the Vietnamese data (discussed in Section 3.3). While the use of Swedish and Vietnamese intersubjective markers shows a clearer orientation towards epistemic perspectives between the speaker and the hearer, the use of the Finnish clitics is rather determined by the interaction intended by the speaker. For the Finnish clitics, in other words, epistemic (a)symmetry is not a triggering factor in the selection of the clitics.

Despite mismatches in functions, subject-person involvement in the Finnish, Swedish, and Vietnamese data show similar distributions. As is summarised in Table 4, markers with either recognitional or interactive functions (Finnish =hAn, Swedish ju, and Vietnamese proximal đây and distal kia/cơ) occur more often with first person on the one hand. Meanwhile, markers with a more declarative nature (Finnish =se, Swedish väl, and Vietnamese medial đấy and ấy/ý) are more common in contexts with second-person involvement on the other.

The comparison in Table 4 illustrates the potential for further language-specific studies, which will ultimately enable a more profound typologisation of intersubjective markers cross-linguistically. For instance, albeit a qualitative analysis on intersubjective uses has also been previously made between the German doch and Finnish =hAn (e.g. Liefländer-Koistinen 1989), it would also be interesting to compare Swedish and Finnish markers of engagement with other Germanic languages as well as other languages with intersubjective markers developed from referential lexemes with such quantitative data.

5.4 Other noteworthy observations from the data analysis

Beyond the two clitic forms =hAn and =se, the questionnaire study does not pay attention to the conjoint ‘double clitics’ =hAn=se due to shortage and unequal distribution of qualitative information provided by the student participants, but as the results of the corpus study show, its uses are even less frequent than those of =se. As shown in Table 3 in Section 4.2, the double clitics are clearly the most common with the second person, as can perhaps be expected based on the highest frequency of =se with the second person. This can probably be explained in a similar fashion through contextual setting as the high frequency of =se with the second person; i.e. it is very appropriate as a response to something someone has just stated (as discussed in Section 5.1). It is also interesting that =hAn=se is very infrequent with the third person. This may suggest that =hAn=se is indeed commonly used as a kind of response to a previous utterance, and in these cases, it most naturally attaches to the second person or first person because third person referents are not necessarily present in the speech event, i.e. the interlocutors’ sphere.

Against our prior assumption, the corpus study provided us with some results coming as counter-expectation. The most striking point is perhaps the occurrence of =se with a plural pronoun: me=se and te=se, as our functional description in Section 3.2 maintained that =se as a less grammaticalised clitic should still maintain its inflectional ability in number. On the one hand, these contexts show a conflict in number agreement, but on the other hand, speak in favour of =se becoming more grammaticalised and gradually losing its ability to inflect when used in the clause-second position.

Moving towards diachrony, the functional dichotomy between =hAn and =se, which becomes even clearer when viewed from a cross-linguistic perspective in Section 3.3, is likely due to a difference in logophoricity of the two clitics, inherited from their referential uses. Namely, the lexical form hän is logophoric, whereas se can also be used exophorically to refer to a subject other than the main activity performer (the difference in logophoricity as was discussed in Section 2.1). From a diachronic perspective, it would be interesting to compare, for instance, the intersubjective uses of indexical and referential elements in Finnish dialects, as well as closely and remotely related Uralic languages to gain more support for the proposed idea that logophoricity has associated effects in the uses of =hAn and =se in the domain of engagement.