Lat vian verbs of speaking and their relations to evidentiality

The paper offers a functional analysis of three Latvian verbs of speaking used in their indicative third person forms – saka, runā and stāsta ‘say(s), speak(s) and talk(s)’ – based on the Latvian language corpus online (www.korpuss.lv) and, additionally, on examples excerpted from Internet discourse. The article discusses semantic and syntactic similarities and differences between these words, the functions of particular constructions distinguished according to specific syntactic criteria (presence vs. absence of a subject), and the use of these verbs in combination with the Latvian verb form traditionally associated with (potential) evidential meanings, the oblique (atstāstījuma izteiksme). While the use of verbs of speaking for introducing reported speech is seen as unproblematic, the relationship between verba dicendi and evidentiality requires more attention and remains in focus throughout this study. In order to offer an indepth overview of the nature of this relationship, the relations between reported speech and evidentiality, as well as the oblique and evidentiality, are also briefly considered. The paper concludes that the possibility of the three verbs functioning as evidential markers depends on considerations of theoretical/terminological nature (how to define evidentiality, should reported speech be considered part of it, etc.) and also, to some extent, on the way the subject is realized in verba dicendi constructions (specific human agent vs. non-specific empty subject and absent or zero subject).


Introduction
Every language has a certain number of verbs of speaking -the so-called verba dicendi.
The group usually includes one or some verbs of the most general meaning -'to say', 'to tell', 'to speak' -and words of more specialized meaning, e.g.: 'to inform', 'to announce', 'to shout', 'to whisper'.Many of such verbs have two meanings at the same time, i.e. the meaning of 'speaking' -using one's vocal organs to produce articulated sounds -and the meaning of doing something to achieve a certain goal (as with 'inform' -sharing some information with other people so that they possess a certain knowledge).Additionally, words such as 'inform' do not always mean 'to speak' -one may also inform using other media, e.g. in writing.This makes such verbs a periphery of the category of verba dicendi (for more on verbs of speaking, cf.Shinzato 2004, Wierzbicka 1987 1 ).This article offers a functional analysis of three Latvian verbs of speaking in their indicative present tense third person forms -saka, runā and stāsta (she/he/it/they 'say(s)', 'speak(s)' and 'talk(s)', respectively; note that the third person singular and plural forms are identical).It is a corpus-based rather than a corpus-driven study, in the sense that the corpus is used to substantiate hypotheses already defined in the literature (Baker 2006, 16, McEnery & Hardie 2012, 5).The examples included here come from the Latvian language corpus (Miljons-2.0 2 ) online (www.korpuss.lv);where necessary, google search has been conducted for more examples of a particular construction.Due to this research design and limitations of language corpora in general and the Miljons-2.0corpus in particular (size, representativeness, balance), the quantitative information given here should be seen as approximate; its purpose is to illustrate and support existing hypotheses rather than generate new ones.
The aim of this paper is to investigate the use of these verb forms in order to answer the following questions: 1. what are the semantic and syntactic similarities and differences between these words?(section 2.) 2. what are the functions of different constructions with these forms, with a special focus on marking evidentiality?Can constructions such as example (1) be considered evidential?(section 3.) 1 While in the literature, the more established and widespread term is 'speech act verbs', I am using the term 'verbs of speaking' in order to avoid associations with speech act theory, which is not in focus here, and to emphasize that the three verbs under study share the semantic meaning of 'speaking', not merely the function of introducing speech acts.
2 At the time of collecting data for this study, the corpus contained ca. 2 million word forms.At present it contains ca.3.5 million word forms.Any numerical references to the corpus (in % or absolute numbers) in this paper relate to the two-million word form version.
[They] also say that a church is buried:obl there and on holy mornings, putting:obl one's ear to the ground, one can:obl hear the bells.Also [they] say that money is buried:obl there.' Except for the general overview of various constructions in section 2, the study mainly focuses on cases of verba dicendi taking complement clauses expressing quoted content.
A complement clause may have the form of an "indirect" quote, in which case it is preceded by the complementizer ka 'that' (as in examples ( 1) and ( 2)).It can also have the form of a "direct" quote, in which case it is not preceded by a complementizer, but typically isolated by quotation marks, and/or introduced by a colon or comma (cf.example ( 7)).
It is being assumed that verbs of speaking (in the third person form) taking a complement clause are unproblematic markers of quoting utterances made by, imputed to, or constructed (cf.Tannen 1986) for another, real or imagined, individual or collective, speaker (here called Other; the exception is when the verbs are used with non-human subjects, section 3.1.2).Put differently, they presuppose the source of knowledge expressed in the complement clause to be Other instead of Self.For convenience purposes, the content of the complement clause will be referred to as 'quote', and this function of verba dicendi as 'quotative', which should both be seen as purely "technical" terms (putting aside discussions concerning the differences between reporting, quoting, citing, etc.); this is the domain of 'reported speech' in the traditional sense of the term as it is used in linguistics, poetics, logic, philosophy etc. (cf.Coulmas, ed. 1986).The term will be avoided here wherever possible to prevent confusing it with 'reportive' evidentiality.
While the quotative function of verbs of speaking is not the focus of this article and thus not problematized, the category of evidentiality is a different matter.The study subscribes to and builds upon the existing literature on evidentiality that defines it as a conceptual category concerned with encoding or marking the source of knowledge expressed in a proposition or the type of evidence on the basis of which a speaker makes a claim (Aikhenvald 2003, Anderson 1986).Latvian lacks the grammatical, obligatory system of evidentiality (in the sense that marking the source or type of/basis for evidence in each utterance is not required as in e.g.Northern Iroquoian or Wintu, cf. Chafe & Nichols, eds. 1986, Dendale & Tasmowski 2001), which is expressed mostly lexically by e.g.sensory verbs and their derivatives (participial forms used predicatively/adverbially).
The oblique, traditionally included in the Latvian mood system and believed to mark reportive evidentiality, should strictly speaking not be considered a mood, and an evidential only partially (section 4; for a discussion, see Arkadiev, Holvoet & Wiemer 2015).Grammatically, it cannot be a mood as it uses participles of the indicative, has a complex paradigm of tenses and freely combines with other verb forms, e.g. the debitive and the conditional 3 .Functionally, its ability to mark evidentiality may depend upon the adopted definition of evidentiality, our understanding of its boundaries as a category/its relations with other categories (especially with epistemic modality), and our interpretation of the functions of oblique constructions.In an earlier publication I suggested that the oblique has three main functions: reportive (when the form encodes type of evidence -hearsay -only), quotative (when the type and specific source of evidence are stated) and citation (when a well-known saying is uttered for illustrative purposes), out of which only the first would be evidential (Chojnicka 2012, 160).
Similar difficulties pertain to the relations between verbs of speaking and evidentiality.
One of the questions concerning the boundaries of this category is whether or not it includes verbs of speaking, and if it does, then which ones -that is, all of them, or only the most conventionalized ones?According to the theory of grammaticalization, the most basic and neutral verbs of speaking in a language are prone to conventionalizing into hearsay evidentials (cf.Chojnicka 2012, 148).If this approach is correct, then what criteria should be employed for accepting or rejecting lexical items?This paper attempts to suggest answers to such questions.
The Latvian examples provided are translated with the goal of expressing the original meaning as closely as possible, even for the price of English translations seeming unnatural or unidiomatic.I have decided against using glosses, with the exception of marking the verbs in complement clauses ind for the indicative and obl for the oblique.It has recently been proposed that Latvian could have two verbal form opposition systems: a modal one (indicative vs. imperative and realis vs. irrealis) and an evidential one (indicative vs. oblique, Holvoet 2007).The full paradigm of both active and passive oblique forms means that for every indicative verb form, there is a corresponding evidential (or 'interpretive ', 2007, 96) form.This approach supports the isolation of the oblique from the Latvian mood system, but I have reservations about seeing the oblique as a grammatical and primary evidential, unless we reconsider the definition and boundaries of the category of evidentiality.
2 Saka, runā, stāsta -semantic and syntactic comparison Saka, runā and stāsta are the most popular verba dicendi in Latvian, which is illustrated by the frequency of their use and the number of possible collocations.The following On the basis of frequency only, the most "basic and neutral" Latvian verb of speaking would be sacīt, which is a conclusion already drawn before (Chojnicka 2012, 148).On the other hand, in that publication saka was considered only partly grammaticalized/ conventionalized into an evidential (p.149).Thus, in the following semantic and syntactic comparison, special attention will be paid to the possible conventionalized/ conventionalizing status of saka as opposed to runā and stāsta.
According to the respective entries in the LLVV dictionary, a) sacīt is used with the following meanings: 'to say (shortly, a text, a word), to express something by talking'; 'to inform by talking, command (i.e.tell somebody to do something)'; 'to have an opinion or think about something and express it in a written or spoken form'; 'to express something with one's actions, activities, appearance, etc.' [as in ko par tevi saka tavs rokraksts?'what does your handwriting say about you?' -J.Ch.]; 'to have such quality that it stimulates certain conclusions, offers information' (LLVV 7.1 (1989)); b) runāt is used with the following meanings: 'to express (something) with spoken language means'; 'to express thoughts, opinions etc. with spoken language means'; 'to express in written form (metaphorically)'; 'to express some information' (referring to a written text); 'to express, mean, reflect' (e.g.referring to a work of art); 'usually 3.pers.; to have such quality that it stimulates certain conclusions, offers information' (LLVV 6.2 (1987); c) stāstīt is used with the following meanings: 'to make known with words'; 'to spread information' (e.g.referring to groups, collectives); 'to make known in written form; to describe in a work of art'; 'to be a source of some content, information' (e.g.referring to text, image); 'to be a sign, indication, evidence; to prove something' (LLVV 7.2 (1991).
These definitions appear to be quite different from each other, suggesting disparities in the use and functions of the three verbs, even specialization (e.g.only sacīt can be used to introduce a command).Such an expectation is confirmed by the findings of the corpus investigation, the results of which are presented in the following subsections (2.1-2.3) and summarized in Table 2 (section 2.4).

Speaking what, speaking how?
The analysis of the use of these verbs in context shows their specializations and differences between their functions in much starker relief than separate dictionary entries.Thus, runā is mostly concerned with the use of language (runāt angļu valodā, nesaprotamā izloksnē, latviski 'to speak in the English language', 'in an incomprehensible dialect', 'Latvian') and with voice -e.g.skaļi, ilgi, daudz ('loudly', 'long', 'a lot').Stāsta is associated with a coherent text, story or narration.The content introduced by stāsta is a narrative, a description of events or a situation (rather than a statement).Saka is associated with words, short and unique acts of speaking, and ritual utterances (saka labrīt 'say(s) good morning').While runā often occurs with adverbs specifying manner of speech, it hardly ever happens with saka or stāsta.
As runā and stāsta are associated with the act of speaking -or speaking as an activitythey can co-occur with other activities undertaken by a subject: (3) Un nav taču nekā slikta, ka Valsts prezidents tagad labāk runā, labāk raksta, labāk dejo.<Mio2, press> 'And it is not bad at all that the State President now speaks better, writes better, dances better.' (4) Saimnieks nenogurstoši stāsta un rāda mapes ar bērnu zīmējumiem.<Mio2, press> 'The host tirelessly talks and shows folders with children's drawings.' Also, stāsta and runā may combine with each other, but they do not co-occur with saka (see (2) and the example below): (5) Sēņu nosaukumi stāsta, sēņu nosaukumi dažādās valodās runā.<Mio2, other> 'Names of mushrooms talk, names of mushrooms speak in different languages.' Both runā and stāsta (although the former much more often than the latter) combine with the preposition par 'about' to express the topic of talk; saka does not combine with par.On the other hand, direct quotes in complement clauses are possible, but not typical for runā, while they are frequent with stāsta and saka.

Speaking to/with whom?
Runā is the most interactional of the three verbs in the sense that it often includes several participants having a conversation, exchanging opinions.Subject may be plural or collective (cilvēki runā, tauta runā 'people say', 'folk say').It does not denote "one way" of processing information: speaker → hearer, but rather a situation where all participants are active speakers and hearers at the same time.It is either multidirectional or non-directional: recipient (addressee) is not important.In fact, in the corpus runā never appears with a recipient (see Table 2).
This contrasts with saka and stāsta, which often occur with a recipient.This means that what matters here is uttering content and transmitting (unidirectional) information.
On the other hand, in constructions involving a recipient in the Dative, but omitting the subject, receiving information (→ hearer) rather than uttering or producing it (speaker →) may be assumed to constitute the pragmatic foreground.
Stāsta seems to specialize in one type of such a construction whereby a subject is missing and the recipient (the first person singular pronoun es 'I' in the Dative form -man) refers to the actual speaker (producer of the utterance).This means that the construction can be translated as 'I hear', 'I am told' and considered an example of introducing hearsay (more in section 3.2).
(6) Man stāsta, ka viens otrs tagadējais politologs esot bijušais zinātniskā komunisma pasniedzējs.<Mio2, press> 'I am told that one or the other current political scientist is:obl a former lecturer in scientific communism.' Saka is also sometimes used with a recipient, but rarely with the subject missing.The construction (subject +) saka + recipient is typical for narrative genres, e.g.: (7) Ella stumj prom svešās sievietes sniegto putras karoti un saka viņai: -Gaudinaties, gaudinaties, gaudinaties.<Mio2, fiction> 'Ella pushes away the spoon of porridge offered by the strange woman and tells her: eat away, eat away, eat away.' In example ( 7), saka introduces a direct speech clause in the imperative (a command).
The possibility of combining saka (but not runā or stāsta) with questions, exclamations or commands suggests semantic bleaching or loss of semantic weight (Lehmann 1995), typical for conventionalization.

Speaking for how long?
Runā is the most "habitual", "continuous", saka -the most "punctual" and "nonrepetitive", while stāsta may be considered to be in the medial position in this respect.
It is a habit of someone to speak in a certain language, style or manner (runā); on the other hand, an event of saying something, of making a statement, happens at a certain point in time and is unique in the sense that one speech act can occur only once (saka).This may be the reason (or one of the reasons) why runā hardly ever occurs with direct speech: direct speech quotes a unique speech act that happens at a certain point of time.
As runā and stāsta denote rather "long" speech events (their derivations prove this point, e.g.runa 'speech', stāstījums 'narrative' vs. sakāmvārds 'proverb, saying'), the quotes that they introduce may be relatively long as well, spanning several sentences in a written text (cf.example ( 11)).
Consider the following examples: in (8), the adverb of frequency nereti 'not rarely' and the adverb of quantity daudz 'a lot' provide evidence for the claim that runā is concerned with long and/or repeated speech events.In (9), runā seemingly introduces direct speech (no complementizer ka 'that'), but it could more correctly be interpreted as emphasizing the repetitiveness of a claim (with its content rather than form foregrounded) and possibly expressing negative (sarcastic) attitude of the speaker (as in English 'keep(s) saying', 'keep(s) claiming').The use of saka here would mean the loss of both effects. (

Summary
The most important findings of the semantic-syntactic analysis of the three verbs are summarized in As the previous section and Table 2 demonstrate, saka, runā and stāsta represent a wide range of constructions and functions.In the present section, these constructions are distinguished according to syntactic criteria in order to investigate functional differences between them, with a special focus on marking evidentiality.The most basic criterion applied here is the presence (3.1) vs. absence (3.2) of a subject.

Human agent
If the subject is present and semantically it denotes a human agent, in most cases it refers either to a specific person (individual speaker) or to a collective (representatives of a profession or association, e.g.ierēdņi 'officials', skolotāji 'teachers'; groups (pusaudži 'teenagers', kaimiņi 'neighbours', vecāki 'parents'), and "empty" subjects, such as cilvēki, tauta 'people, folk'.With "empty" I mean that these nouns are semantically weak and their use appears to be motivated by grammatical requirements, as in 'they say' in English).Each of these types of subjects seems to have a specialized verb -stāsta for specific, individual speakers and runā for collective or empty subjects (although these specializations are to be understood as tendencies rather than "rules").
Stāsta specializing in subjects denoting specific speakers might be a genre effect.In the corpus, such constructions appear almost only in excerpts from press.In most cases, they are concerned with presenting an opinion of an expert -a concrete person who is known, whose name, position, field of expertise are explicitly stated.Such a presentation of the person speaking can be very long, and is supposed to justify the inclusion of a quote: ( The complement clauses of these constructions are usually in the indicative (also possible are participles without auxiliary 4 ), less often -in the oblique.In turn, the oblique is a frequent choice for the sentence that follows directly afterwards, if the quote continues: (11) Skrundeniece Agita Gorsvāne stāsta, ka ārstu apmeklē reti, biežāk pie dakteres Māras Ostašovas aiziet mamma.Mediķi un pacienti esot pazīstami, tāpēc attiecības ir citādākas nekā pilsētās.<Mio2, press> 'The resident of Skrunda, Agita Gorsvāne says that she rarely visits:ind a doctor, it is her mother that more often goes:ind to the doctor Māra Ostašova.Doctors and patients are:obl familiar [with each other], so their relations are different than in cities.' Since the source is known from the immediate context, the function of the oblique here is quotative rather than reportive (no evidentiality).
Stāsta is also preferred for introducing direct quotations, which are usually attributed to a specific speaker.In the corpus, almost half of tokens of stāsta are used to introduce direct quotes, which constitutes most significant difference between semantically "close" stāsta and runā (the latter being dispreferred for direct speech).
It may be mentioned that the association of stāsta with "long talking" and competent, expert speakers (subjects) produces a possible ironic effect when the complement of stāsta is short, and the quoted speaker mistrusted, as in the following example: (12) Mērs pēc šāda aicinājuma smaida vien un stāsta, ka Rīgas domes koalīcija esot stabila.<Mio2, press> 'After such an invitation the mayor just smiles and says that the Riga City Council coalition is:obl stable.' Here, the function of the oblique is completely different than in example ( 11) and comprises expressing the speaker's distance (cf.section 4).
When it comes to runā, the corpus offers only two examples with a complement clause and a human agent as subject, and they both involve a collective subject (in both cases, the complement clause is in the indicative), e.g.: (13) Psihologi runā, ka mode uz dabīgo kosmētiku ir pasakās minētā dzīvā ūdens un atjaunojošā ābola mūsdienu variants.<Mio2, press> 'Psychologists say that the fashion for natural cosmetics is:ind a contemporary counterpart of the live water and restorative apples found in fairy tales.' Next to these two examples the corpus contains 33 cases of runā, ka without a subject (3.2), suggesting a strong preference for this variant.Additionally, the results of a google search complementing the corpus analysis show that, while occurring most frequently with zero subjects, runā also specializes in non-specific, semantically weak human subjects such as cilvēki, tauta, ļaudis 'people, folk', daži 'some', and even metonymic pasaule 'the world'.
As an exception, runā combines with subjects denoting specific human agents when there is no complement clause: (14) Dāmas un kungi, šeit runā jūsu kapteinis.<Mio2, fiction> 'Ladies and gentlemen, here speaks your captain.' The focus here, however, is on the act of speaking rather than its content, and such constructions never function to mark reported speech or hearsay.
In most cases, saka takes a complement clause in the indicative, followed by the oblique (about one-third of the examples), participle without auxiliary, and the imperative (for a discussion, see section 4).
All the examples discussed in this section, with the exception of example ( 14), are cases of indirect speech.Whether such constructions (human subject + saka/runā/stāsta + complementizer ka) belong to the category of evidentiality depends on our understanding of the relation between reported/indirect speech and evidentiality, about which there is no agreement in the linguistic community.Some researchers consider reported speech as one of the ways of expressing evidential values.For instance, Li says that "direct quote and indirect quote are forms of evidentiality" (1986,41).Mushin (2000, 929) also believes that quotation is used to mark evidentiality, but not in all cases.For languages without specialized evidentials, Haßler (2002, 159) allows modal verbs, past tenses, adverbs and reported speech markers in the evidential function (although probably neither Mushin nor Haßler has in mind a construction with a realized subject denoting a specific human agent).
I tend to agree with Aikhenvald (2004) that reported speech and evidentiality are related, but distinct linguistic phenomena.While evidentiality is a conceptual category, reported speech is a set of devices used to retell utterances -"constructed dialogue" (Tannen 1986).
There is definitely a difference between saka, runā, stāsta constructions with subjects denoting specific human agents and non-specific or "empty" subjects.While the former are prototypical forms of reported speech and not evidentials, the latter are conceptually "closer" to evidentiality, even if Wiemer (2010) considers constructions such as People say that P "trivial" markers of hearsay.
A note on direct speech: the complementizer ka 'that', present in most cases above, marks the choice of indirect over direct speech.All the three verbs may introduce direct speech (although the use of runā in this function is rare), e.g.: (17) Vēlētājs bieži saka: balsojot par "mazo" vai nezināmo partiju, pazaudēšu balsi, jo šo partiju tik un tā neievēlēs Saeimā.<Mio2, press> 'A voter often says: voting for a "small" or unknown party I will lose my vote, as this party will not be elected into the Parliament anyway.' It must be emphasized, however, that utterances such as in ( 9) and ( 17) represent "direct" speech only in a stylistic sense: neither retells what a specific, concrete speaker has uttered before.Rather, they construct, impute a "typical" statement that an imaginary "typical" company representative (in ( 9)) or voter (in ( 17)) might give.For this reason, Tannen's term "constructed dialogue" is much more accurate than "direct speech" to describe formulations such as these.

Non-human agent
There are some examples of the use of saka, runā and stāsta with a subject denoting a non-human (and predominantly non-animate) agent (teksts 'text', bilde 'picture', statistika 'statistics', also figurative use: sirds 'heart', māja 'home', etc.), although the relative frequency of such constructions is not high: about 10% for runā, less than 1% for stāsta and saka (another significant difference between runā and stāsta).However, if the construction non-human agent + verb of speaking is followed by a complementizer ka and a complement clause, then the verb tends to be saka, e.g.: (18) Tā ir laba zīme, Ieva automātiski nodomā, suns saka, ka mans ceļš ir pareizs.<Mio2, fiction> It is a good sign, Ieva thinks automatically, the dog says that my way is:ind correct.
This could be considered evidence for the semantic bleaching of saka -a property of conventionalization.
In contrast to constructions involving a human agent as subject, constructions involving a non-human agent cannot be considered as marking reported/indirect speech.Here, verba dicendi are rather used to mean 'show', 'denote', 'indicate', or 'prove'.In example (18), the behaviour of Ieva's dog "tells" her (proves) that she is heading in the right direction.In ( 19), previous experience functions as a basis for making a statement.In both cases, then, saka could be interpreted as pointing to basis of knowledge or evidence for knowledge.
Introducing the concept of epistemological positioning (EP), Bednarek suggests a distinction between a source and a basis for knowledge.When it comes to sourcing, utterances may be divided into averrals (the source is Self) and attributions (the source is Other).In turn, the basis for knowledge may be further distinguished into the following categories: perception, general knowledge, proof, obviousness, unspecified, hearsay, mindsay (2006,644).This means that if it is accepted that the category of proof may be represented by verba dicendi with non-human subject such as in the examples above, these constructions are to be considered evidential markers.
Additionally, while they are not cases of attribution (the source is Self), these constructions definitely "give some sort of indication about the evidential basis of the writer's averral" (ibid., 647), which Bednarek also considers a function of evidentiality.
It may be mentioned here that within Bednarek's framework, attribution is also an evidential function (including "trivial" markers of hearsay such as people say, they say and I've been told, p. 654), which would make constructions discussed in section 3.1.1part of evidentiality as well.
Still, it should be kept in mind that Bednarek's proposal is hardly compatible with more central and widely accepted approaches to evidentiality, such as Aikhenvald (2003Aikhenvald ( , 2004) ) or, more relevantly for Baltic languages, Wiemer 2010.Within these approaches, verbs of speaking with non-human agents (based averrals) and attributions would rather be considered descriptive or trivial means of referring to the source or base of knowledge.On the other hand, it can be argued that based averrals differ from "trivial" constructions such as cilvēki saka in some important aspects: verbs of speaking are used here with a figurative meaning (personification of non-human/inanimate objects), they presuppose a more active role of the speaker than in a "typical" quotation (understanding, interpreting, summarizing information instead of just repeating it), and the subject cannot be ellided without a change of meaning.

Subject absent
The frequency of verbs of speaking without a subject has turned out lower than expected.The most frequent in this construction is runā (34 tokens, or ca.7.5% of all examples of this verb), followed by saka (19 tokens, or ca.1.5%) and stāsta (5 tokens, or only 0.5%).
Many constructions with stāsta and zero subject are cases of ellipsis (the subject is retrievable from the immediate context), and so differ significantly from constructions with runā and saka where the subject cannot be retrieved ('they say' -general hearsay).Additionally, a look at the Internet discourse suggests that when the subject is missing, stāsta tends to combine with a recipient -usually in the first person and always with the complementizer (man stāsta, ka), cf. ( 22) below.

With complementizer
In most cases, zero subject constructions combine with the complementizer ka and a complement clause.Here, there seems to be a significant difference between saka on the one hand, and runā and stāsta on the other hand.This difference pertains to the verb form in the complement clause: while saka most often combines with a complement clause in the indicative (cf.example (1)) or participle without auxiliary (example (21) below), runā and stāsta combine with complement clauses in the oblique in the majority of cases (for stāsta in 4 out of 5 examples): (20) Runā, ka vīriešiem esot nosliece uz poligāmiju, bet man jau sen likās, ka šai ziņā sievietes no mums daudz neatpaliek.<Mio2, fiction> '[They] say that men have:obl a tendency towards poligamy, but has long seemed to me that in this sense women are not far behind us.' (21) Saka, ka īstam vīrietim dienā jāapēd vismaz viens mango.<Mio2, press> '[They] say that a real man should eat at least one mango per day.' (22) Man stāsta, ka patlaban viņiem notiekot audits, tāpēc nevarot izmaksas veikt, būšot pirmā ceturkšņa beigās, bet nu ... diez kā neticas :( <http://www.liepajniekiem.lv/forums/ekonomika/european-prosperity-club-76188> '[They] tell me that they have an audit happening:obl right now, that is why they cannot:obl make payments, they will be:obl there at the end of the first quarter, but... somehow I do not believe [it] :(' Examples ( 20)-( 22) illustrate an important property that goes hand in hand with the indicative vs. oblique distinction: runā and stāsta tend to combine with the oblique in contexts of contrast (note the use of the conjunction bet in (20) and ( 22)).The speaker distances her/himself from the content of the quoted proposition and counters it with another one which, according to her/him, is more accurate.Saka does not have this effect: there are no examples of saka in context of contrast in the corpus.It may be concluded that saka marks the speaker's support for or belief in the content of the quoted proposition.This may also explain the use of saka for based averral or proof, as in examples ( 18) and ( 19).Also, only saka can combine with such modifiers as ne velti ('not without reason'), example (26) below.These findings suggest that it is not unjustified to consider saka a marker of speaker support and not merely neutrality (lack of distance).Saka could then be "neutralized" by using the oblique in the complement clause.
It may also be proposed that runā specializes in contexts of contrast and lack of speaker's support when the original source of knowledge is irretrievable (general hearsay), while stāsta -when the source is known, also when ellided (compare examples ( 12) and ( 22)).
It should be kept in mind, however, that these uses may be restricted to or excluded from specific genres.For instance, example (2) illustrates the use of both runā and stāsta with the oblique, within one coherent text fragment.But this fragment comes from a narrative genre with very strong conventions (legend, myth, fairy tale).In this case, the use of runā/stāsta together with the oblique should not be considered as a marker of speaker's distance or disbelief, but rather a stylistic artefact of the narrative genre.

Without complementizer
There are only four examples of saka and two of runā followed by a complement clause without a complementizer.Excluded from this group are cases such as the following: (23) Stāsta Lolita Krēmere: "Darba pamatideja -svešinieku nekur nepieņem."<Mio2, press> 'Lolita Krēmere says: "The basic principle of work -a stranger is not accepted anywhere."' Here, the position of stāsta at the head of the sentence does not imply a zero subject, but an inversion, and the lack of a complementizer is justified by the use of "direct" speech (recall example (9)).
Let us now consider these examples of saka and runā without complementizer: ( In all these examples except for (26), the event of speaking is backgrounded, irrelevant -saka and runā are relatively free syntactically, providing meta information concerning the (reported) proposition, which is foregrounded.Still, they are not fully independent as their position is fixed at the top of the sentence.It may be assumed that these markers are currently undergoing the process of conventionalization (which is not to say that we should expect this process to ever be completed).
When it comes to marking evidentiality, I would include the constructions with "real" (irretrievable) zero subjects in the category of hearsay, while the cases of ellipsis should be considered similar to the constructions with subjects denoting human agents: indirect speech if the retrieved subject denotes a specific speaker; somewhat closer to evidentiality if it is a collective, and even closer if it refers to "people in general" ('they say'), see also section 5.
4 Saka, runā, stāsta and the oblique Throughout the article I have provided numerous examples for the use of saka, runā and stāsta with complement clauses in the oblique ((2), ( 6), ( 11), ( 12), ( 16), ( 20), ( 22), ( 25), ( 27)).In this section, I would like to analyze these cases from a slightly different perspective -taking the oblique, not the verbs of speaking, and its potential to mark evidentiality as the point of departure.
The first conclusion that comes to mind when considering the examples listed above is that there seem to be two groups or types of utterances containing the oblique -one whereby the speaker's doubt, distance, uncertainty or lack of support is expressed, and another whereby it is not (utterances neutral in this respect).In the "neutral" group there are cases in which the use of the oblique seems to be a genre-related or stylistic artefact (example (2)).There are also such examples as ( 11), where a sentence in the oblique is separated from the sentence that contains the verb of speaking (introduction of the quote).
Here, the oblique carries a kind of disambiguating function, marking the sentence as belonging to the quote (continuation of it) and not as the speaker's own, original words.It is a quotative equivalent of the indicative -a stylistic choice without evidential or epistemic overtones (Chojnicka 2012, 158).On its own (without the preceding sentence), the sentence in the oblique could be considered to mark reportive evidentiality (encode the type of evidence as hearsay).
I would also argue to include examples ( 16) and (25) in the "neutral" group.If the hypothesis that saka is inclusive (presupposes the speaker's support for or belief in the content of the proposition) is correct, the oblique form of the complement clause may function to cancel this inclusive, positive value of saka, to neutralize it.
The other group includes cases of contrast (( 20) and ( 22)) and sarcasm (( 12)).Here, the use of the "neutral", non-inclusive verbs of speaking stāsta and runā makes it possible for the oblique to mark negative attitude of the speaker.Such uses invite discussion concerning the possibility of the oblique to mark epistemic modality, especially if this negative attitude is interpreted as doubt or lack of certainty.In any case, this epistemic function could only be a pragmatic extension and not a basic meaning of the oblique.
Where contrast, sarcasm or another means of marking the speaker's negative attitude cannot be ascertained, but where runā or stāsta are used, as in examples ( 6) and ( 27), the oblique could work as a marker of distance -not an "actively" negative attitude, but not a neutral one either.Overt marking of distance should be seen as a move away from (unmarked) neutrality towards lack of belief.
The above considerations suggest that the use of the oblique in complement clauses of verbs of speaking is always grounded and never redundant -the oblique does add meaning overtones to the utterance that are not expressed by these verbs and that extend beyond its use as a stylistic quotative equivalent of the indicative.However, these meaning overtones are never evidential, since the source (specified in case of realized subject) and type of evidence (hearsay) have already been provided by the verb (and this regardless of whether we consider verbs of speaking to be markers of evidentiality or not).The use of verbs of speaking, with or without subject, already states both the source (Other) and type of or basis for knowledge (hearsay).The oblique can only mark reportive evidentiality if it is the only marker of evidence or basis for knowledge in its immediate context.
I believe that marking evidentiality involves encoding the information about the type of source or basis of knowledge (for example, hearsay or visual), rather than naming this source or basis explicitly with lexical means (for example, 'X said that…', 'I saw that…').
In a similar vein, the category of tense encodes the meaning of present, past or future in general terms, while more specific information concerning the timing of an event must be named with lexical means (e.g.'yesterday', 'next Sunday').Naturally, specifying the timing does not cancel the tense meaning of the verb 5 .However, an important difference between tense and evidentiality in Latvian is that the former is obligatory while the latter is not.Another difference is that, as I believe, coding evidentiality and explicitly naming a source of knowledge are decisions that have diverging motivations and effects; the latter does not simply provide more specific information to complement the former.Marking the information as not originating -i.e., pointing away -from the speaker without specifying the source (because, for example, it is unknown or irrelevant) is different than explicitly naming, i.e. pointing towards, the source.
In fact, there have been numerous attempts in the literature to find alternative explanations for markers commonly considered evidential, motivated by the fact that these markers seem to have in common the property of pointing away from the source of knowledge (i.e., the speaker) rather than pointing towards it.One such attempt is the theory of territory of information, originally developed for Japanese (Ishida 2006).Japanese uses direct verb forms to express what falls within the domain of the speaker's territory of information, and indirect forms for information falling outside this domain.
Another similar approach is to treat these markers as deictic operators.Faller (2004) proposes to interpret the Quechua marker -sqa in such terms: as a deictic, rather than an evidential, it can be seen as placing the proposition outside the perceptual field of the speaker (Chojnicka 2012, 154).
Considering the Latvian oblique as a marker of indirectness or deixis does explain why it can be used for proverbs, sayings, legends, myths, folk beliefs, etc., for constructed, hypothetical, or imputed utterances, and even for reporting one's own dreams, while it is never used for self-quotation.In my opinion, also the fact that the oblique is never used in constructions with non-human subjects (as discussed in section 3.1.2)justifies this approach: formulations such as teksts saka 'the text says' presuppose that the speaker has first-hand access to the evidence or proof of knowledge provided by the text.It should also be emphasized that the discussion above is not meant to suggest that the oblique can never be evidential, just that evidentiality is not its only, or even its most frequent, function.

Conclusions
On the basis of considerations in the previous section, I am proposing not to treat forms of the oblique in complement clauses of verbs of speaking as markers of evidentiality.They can be stylistic equivalents of the indicative required by the genre or for disambiguation purposes, or markers of speaker's distance down to more negative attitudes, possibly including epistemic values.All these possibilities can be described in terms of being outside of the speaker's perceptual field, which seems to be the only value of the oblique suggested in the literature so far that is present in all its uses.In this interpretation, the oblique is primarily an indirect marker, fulfilling the evidential function if there are no means pointing towards the source of information in its immediate context.
What about evidential meanings of the verbs of speaking themselves?I would like to suggest conceptualizing possible functions of the verba dicendi discussed here along a cline with reported speech (quoting) on one end and evidentiality on the other end.Here, evidentiality is not a discrete category, but rather a matter of degree.This conceptualization has allowed me to refer to different constructions as being "close" to evidentiality (i.e., closer than to reported speech) earlier in the article.
Although I organized my argument according to syntactic criteria -the first-level distinction was present vs. absent subject -semantic and pragmatic features have turned out to play a more important role in arranging different constructions along the cline.Thus, verbs of speaking with a specific human agent as subject, or a missing subject that can be traced to a specific human agent, are prototypical reported speech constructions.
Verbs with collective subjects or missing subjects retraceable to collectives or groups are located around the middle of the cline.Verbs with empty subjects (e.g.cilvēki, ļaudis) and with zero/irretrievable subjects are closer to evidentiality than to reported speech.And the relatively independent verbs lacking a subject and a complementizer are on their way to being conventionalized into reportive evidentials, although there is no way of saying if this process will ever be completed.
Finally, in an interesting turn, the three verbs of speaking in their figurative uses -uses that do not denote 'speaking' at all -combined with non-human, non-animate subjects, mark based averrals.However, Bednarek's proposal is still poorly integrated in better known and widely accepted frameworks represented e.g. by Aikhenvald (2003Aikhenvald ( , 2004)), which is a problem waiting to be addressed in future studies. 3

Table 1 .
table presents the number of tokens found in the corpus for each of the verbs in question: The frequency of saka, runā and stāstā in the Miljons-2.0corpus and opinions, to replace the routine press release in which a representative of a company says how good [it is] that there are:ind farmers, and how good [it is] that there are:ind farmers' markets, and pretends to be the people's savior, but whatever.' 8) Zemkopības ministrija nereti daudz runā, kā zemniekiem dzīvot, -vai piena ražošanas pašizmaksai jābūt 19 vai 20 santīmus lielai.<Mio2, press> 'The Ministry of Agriculture not rarely says a lot [on] how farmers should tautas glābējiem, bet tas nu tā.<Mio2, press> 'Still, we do have farms with a name and a status that we should praise sometimes, we do have reasonable-minded and intelligent farmers and craftsmen who can share their thoughts

Table 2 .
Semantic comparison of the three verbs