Information structure of converb constructions: Estonian -des, -mata and -maks constructions

2.1 The semantics of Estonian converb constructions

The tense of Estonian converb constructions is relative to the event of the main clause. There are three unmarked relative tense converb constructions in Estonian: the -des, -mata and -maks constructions.

The -des construction (1) expresses an occurring event, while the -mata construction expresses an event that does not take place (2). The event of -des and -mata constructions is usually simultaneous with the event of the main clause. However, depending on the context, the event of -des and -mata constructions may also be understood as (immediately) preceding or following the event of the main clause (Erelt 2017c: 814).

The -maks construction (3) is a converb construction serving as an adverbial of purpose which was intentionally devised in the early 20th century. It is generally regarded as stylistically marked and it is subject to syntactical limitations and primarily found within specific registers (Uuspõld 1980). The -maks construction serves primarily as a (non-obligatory) adjunct (Erelt 2017c: 799). In Estonian, the primary obligatory non-finite purpose clauses are the -da infinitive and -ma infinitive constructions (Erelt 2017c: 780–782, 792–799). In contrast to the -des and -mata construction, the -maks converb construction always expresses a posterior event, specifically, an event that (hypothetically) follows the event of the main clause. Given that posteriority is the sole temporal relationship with respect to the meaning of purpose, I also consider the -maks construction as an unmarked tense construction.

In addition to the three unmarked tense converb constructions, Estonian features two marked tense converb constructions, the -nud and -tud constructions, which express an event that precedes the event of the main clause. The -nud and -tud constructions have been considered as anterior tense counterparts to the -des construction (Erelt 2017c: 807). For the purposes of this study, I exclusively concentrate on the unmarked tense (-des, -mata and -maks) converb constructions, excluding the marked tense (-nud and -tud) constructions. This approach allows for a more focused examination of the information structural effects associated with the diverse adverbial semantics (and polarity) found in converb constructions.

Converb constructions express an event that serves as a circumstance of the event of the main clause. The semantic function of the converb construction depends on the (e.g. temporal, causal) nature of this circumstance. The semantic function of converb constructions has been described using a variety of categories both in typological studies (Kortmann 1991; König 1995; Nedjalkov 1995) and in studies of converb constructions of one language family or one language (Killie and Swan 2009; Nedjalkov 1998), including Estonian (Erelt 2017c; Plado 2015a, 2015b; Simmul 2018; Valijärvi 2003; Veismann et al. 2017; Uuspõld 1966).

The semantic polysemy of -des and -mata constructions is extensive (Erelt 2017c: 815–818). Estonian converb constructions can have meanings characteristic of both clauses (e.g. result, specification) and phrases (e.g. means, manner). In this article I makes use of typologically known semantic categories to describe the polysemy of converb constructions (e.g. time, concomitance, manner, cause, condition; see Kortmann 1991; König 1995; Erelt 2017c: 815–818). Among these categories, I distinguish between basic and complementary functions, based on the previous description of the polysemy of -des and -mata constructions (see Simmul 2018). For the sake of explanatory power, I have somewhat generalized this system for this study. Next, I describe the semantic categories used in this study.

Converb constructions always carry their basic function, which is either time or concomitance. A construction carrying the basic function of time expresses the temporal context of the event described in the main clause (1, 4), while a construction carrying the basic function of concomitance is itself placed within the context of the event described in the main clause (2, 3, 5). The -des construction, having the most extensive polysemy of Estonian converb constructions (Simmul 2018: 863–865), can carry the basic function of either time or concomitance, the -mata and -maks construction always carry the function of concomitance as they do not express the temporal context of the event of the main clause.

The difference and opposition between time and concomitance is illustrated by examples 4 and 5. In example 4, the converb construction with the function time serves as a temporal context of the event of the main clause. In example 5, the event of the construction with the function concomitance is placed within the temporal context of the event of the main clause.

In addition to the basic function, a converb construction can also carry a complementary function, based on a possible additional causal, descriptive or contrastive relationship between the two events. In this article I describe the semantic function according to categories based on combinations of basic and complementary functions. A total of 15 combinations occur in the data: time-none^[3] (4), concomitance-none (5), time-means (6), concomitance-means (7), concomitance-cause (8), concomitance-condition (9), concomitance-manner (21), concomitance-concession (22), concomitance-purpose (32), concomitance-result (33), concomitance-contrast (34), time-cause (40), time-concession (44), time-condition (47), concomitance-specification (48).

2.2 The word order of Estonian converb constructions

There is no clear unmarked word order position for Estonian converb constructions. As an adjunct, the converb construction would be expected to be positioned inside the main clause, typically as the third constituent, after the subject and predicate (example 39; Sahkai 1999: 29). As a non-finite clause, however, converb constructions tend to be located further in the clause than would be expected on the basis of syntactic constituency (Sahkai 1999: 32). Most frequently, Estonian converb constructions are indeed postposed to the main clause (Simmul 2020: 228), but preposed and interposed constructions are also regular, so the word order of converb constructions in Estonian is highly variable (Lindström 2017: 560–561; Remmel 1963: 292–298; Simmul 2020). The variation of the position of converb constructions is related to the semantic function, the position of the converb, the length of the construction (Simmul 2020: 237) and the information structure (Simmul 2021: 327). In addition to this, the position of the converb also varies. The converb can appear at the beginning or the end of the construction, can function as a construction by itself, or, more rarely, can occur in the middle of the construction.

3.1 Information unit

This article’s treatment of information units is rooted in Halliday’s concept of information units (1967, 1985) and Chafe’s concept of idea units (1979, 1985, 1992). An information unit is a linguistic unit that connects a (new) piece of information to a (known) piece of information, effectively asserting something novel and contextually relevant (see Grice 1975: 45–46). The new information within a unit aligns with what Lambrecht (1994: 52) defines as assertion. Meanwhile, the known or readily accessible information, as described by Chafe (1987), corresponds to the concept of presupposition according to Lambrecht.

The amount of asserted and presupposed information within an information unit is organized according to the categories of quantity, relation, and manner (see Grice 1975: 45–46; Halliday 1985: 275). Assertion renders an information unit effective and worthy of expression, but in excess it impedes comprehension, thus contradicting the category of manner. On the other hand, presupposition ensures that the information unit remains contextually relevant and intelligible, yet excessive presupposition goes against the category of quantity (see Grice 1975: 46).

An information unit comprises information elements, which are smaller linguistic units that express distinct pieces of information, collectively forming a complete idea within a given context. Although an information element can be intricate both semantically and grammatically,^[4] it operates as a cohesive conceptual entity within the information unit. I differentiate five types of information elements: focus, background part of the comment, topic, frame, and prominent element. In the following sections, I will delve into these categories, drawing insights from the framework of Estonian information structure descriptions.

3.2 Information structure of Estonian

The structure of Estonian information units has been mainly described along three interacting dimensions, i.e. topic-comment, background-focus, and known-new. These information-structural dimensions have in turn been associated with the prosodic, syntactic and semantic domains (e.g. Asu et al. 2016; Lindström 2005; 2017; Tael 1988).

The topic-comment dimension is the aboutness relation of the information unit: the comment gives information about the topic which, in turn, states the subject matter of the comment and thereby makes the comment relevant by linking it to some accessible piece of information (Klumpp and Skribnik 2022: 1019; Lindström 2017: 537–539; Reinhart 1981).

The focus is the element that expresses the most unpredictable piece of information in the information unit. In the description of the information structure of Estonian, the focus has been mainly treated as the one most unpredictable element, i.e. narrow focus (e.g. Valin and LaPolla 1997), argument focus (Lambrecht 1994; 2000) or constituent focus (Lindström 2017: 545). In description of Estonian, the so-called wide focus or predicate-focus (Lambrecht 1994: 228) has been mainly treated as the comment: the part of the information unit that provides new information about the topic (Lindström 2017: 537, 545).

The dimension of known-new is about how familiar and accessible a piece of information is in the speaker’s assessment to the addressee (cf. Prince 1981, 1992; Chafe 1987; Lindström 2017: 542–544).

In description of Estonian, these information-structural dimensions have been associated with word order, lexicogrammatical markers, length of linguistic units and the number of modifiers (e.g. Lindström 2017; Remmel 1963; Tael 1988). The order of syntactic constituents, i.e. word order in Estonian is syntactically flexible and information-structurally functional. A typical information unit begins with a known topic and ends with a new focus. Between them is the background (i.e. unfocused) part of the comment, which relates the focus to the topic (Lindström 2017: 549–551). Alongside information structure, Estonian word order is regulated by a syntactic tendency, the so-called V2 rule, according to which the predicate in simple declarative sentences and main clauses tends to appear in the second position (see Lindström 2017: 547, 549, 551; Sahkai and Tamm 2019; Tael 1988: 40). The combined effect of information-structural considerations and the V2 rule is that a typical Estonian information unit features a known subject in the topic role in initial position, followed in the second position by a predicate in the background of the comment, and then concludes with a new focused modifier (Asu et al. 2016: 181–183; Lindström 2017: 545–549; Tael 1988). Thus, the main constituent order of Estonian is SVX on the syntactic level, and Top-BGoC-Foc on the level of information structure (10).

In Estonian, lexicogrammatical markers mainly contribute to indicate those information-structural properties of a linguistic unit that are unusual for its word order position (vt Lindström 2017: 544). If the topic expresses a new piece of information or if the comment contains known pieces of information, the word order usually adheres to the topic-comment structure (topic before comment). The somewhat atypical knownness/newness is thereby encoded lexicogrammatically (Lindström 2017: 543). If the referent of the topic is new (example 11) or if the referent of the comment is known (12), it is often marked by a pronoun expressing the (in)definiteness of the referent. Also, lexicogrammatical markers are used to highlight a focus that is not information unit-final (13).

Knownness and newness are also related to the number of words and modifiers of linguistic units. Linguistic units expressing new information tend to contain more words, to be more often modified and to have more modifiers than linguistic units expressing known information (Lindström 2017: 544, 547, 550).

In addition to the main information element order (Top-BGoC-Foc), Estonian has another regular element order, which starts with the so-called “alternative element” usually followed by predicate, topic and focus (Lindström 2017: 539, 551–552; Tael 1988: 6). In the description of Estonian, this alternative element has been called a secondary topic (Lindström 2017: 551–552). In this study, I avoid the term secondary topic to distinguish the alternative element from the aboutness-topic. The reason for this is that, even in the case of the alternative element order, the comment is not mainly about the alternative element, but still about the aboutness-topic, despite it being not positioned at the beginning of the information unit (Lindström 2017: 539).

Among those “alternative elements”, I distinguish between frame and prominent element. The frame is an information unit-initial element that contextualizes the rest of the information unit (see Chafe 1976: 50–51; Jacobs 2001: 655–657; Krifka 2007: 35–37; Krifka and Féry 2008: 129; Tael 1988: 31). A typical frame is a localizing adverbial: a locative place or state adverbial, an adverbial of condition, or a localizing time adverbial, that places the rest of the information unit in a spatial, causal or temporal context. The information unit in (14) is temporally contextualized by the frame Reedel, ‘on Friday’. At the same time, the comment is still understood to give information about the topic ta, ‘he’.

In Estonian, the frame does not have the aboutness relation with the comment and does not interfere with the aboutness relation between topic and comment (Lindström 2017: 539). With this in mind, unlike some approaches (e.g. Chafe 1976; Erteschik-Shir 2007), I do not treat a frame as a type of topic, but as a distinct information role.

According to the description of Estonian information structure, the alternative element, including the frame, is always the first element of the information unit (Lindström 2017: 539). However, the information unit-initial position in itself is not a sufficient criterion by which to define or identify the frame nor any of the information roles. Depending on the information unit-internal relations, the initial element may have any information role: topic (11), background part of comment (15), focus (16) or prominent element (17).

In addition to the frame, an alternative element may have the information role of prominent element (Tael 1988: 38–39): an unfocused emphatic element that does not express the most unpredictable piece of information in the information unit but is emphasized relative to another element in the information unit (usually the background part of the comment). Unlike a frame, a prominent element is placed at the beginning of the information unit not for the sake of contextualization, but for the sake of emphasis. Prominent element as a distinct information-structural category makes it possible to analyze an information unit with multiple emphasized elements as including one main, i.e. the most unpredictable focus while the other emphasized elements function as prominent elements. In (17) the prominent element is muretult, ‘carelessly’ which is emphasized by its information unit-initial position. The focused piece of information is expressed by the information unit-final element kingadele, ‘on shoes’.

The information structure of Estonian information units exhibits variation. Information units can feature different combinations of information elements and are not required to include all of the elements. In an information unit that contains a focus, background part of the comment, topic, and frame, the division of roles can be described as follows.

The focus conveys the most unpredictable piece of information. The background of the comment offers an immediate context for the focus and, in conjunction with the focus, forms the comment. The topic contextualizes the comment according to the aboutness relation. The frame contextualizes the rest of the information unit, encompassing the entire topic-comment structure. An example of such an information unit is presented in (14), and the structure of this type of information unit can be depicted as follows:

Furthermore, the background of the comment, the topic, and the frame collectively constitute the background of the information unit, representing the portion of the information unit that is backgrounded relative to the focus.

The elements of the same information unit can be interconnected through three types of information unit-internal relations. Namely, these relations are the comment-forming relation between the focus and the background part of the comment, the aboutness relation between the comment and the topic, and the contextualizing relation between the frame and the rest of the information unit. These relations serve to connect the elements of an information unit in such a way that they collectively express a coherent and contextually relevant idea.

4.1 Data

This paper is based on a corpus data set consisting of sentences that contain -des, -mata or -maks constructions. The sentences with -des and -mata constructions come from the 1990s print media (NEWS; 865,000 words) and fiction (FIC; 602,000 words) subcorpora of the Corpus of Written Estonian. I started collecting the data from the -mata constructions, the amount of which is the most difficult to estimate based on the initial results of corpus inquiry. This is primarily because the -mata form is multifunctional. It functions not only as a converb but also as a participle and part of compound predicates. Furthermore, the -mata form is relatively frequent in lexicalized and grammaticalized converb constructions. I collected all the -mata converb constructions occurring in the corpora (1,255). Then, I removed the lexicalized and grammaticalized converb constructions (e.g. kogemata, ‘accidentally; lit. without experiencing’; kahtlemata, ‘undoubtedly; lit. without doubting’; tingimata, ‘necessarily; lit. without bargaining’), resulting in 476 -mata constructions in the data. Example (19) illustrates a concessive postpositional phrase that has been grammaticalized from the -mata converb construction.

Collecting -des constructions, I had in mind that data should reflect the fact that -des converb constructions are more frequent than -mata converb constructions. I also wanted the data to represent the rarer semantic functions of -des construction. On the other hand, I wanted a comparable number of -des and -mata constructions. With that in mind, I collected approximately twice as many -des constructions as -mata constructions, collecting -des constructions from a fifth of the text from which I had collected -mata constructions. There were 1,025 -des forms in the selected text. After removing the lexicalized and grammaticalized constructions (e.g. alates, ‘since; lit. beginning’; möödaminnes ‘passingly; lit. passing by’; ausalt öeldes, ‘honestly; lit. honestly saying’) 825 -des constructions remained.

Given the rarity of the stylistically marked -maks construction in fiction, I collected -maks constructions from the larger Balanced Corpus (BC), which comprises 15 million words encompassing various registers, including scientific, print media, and fiction. I collected 500 -maks constructions, resulting in a dataset containing a total of 1,803 converb constructions.

The corpora also provided the opportunity to consider the context of the sentences containing the converb constructions. Assessing the converb constructions and their relationship with the main clause in their context, I systematically annotated the values of the 11 information structural, semantic, morphosyntactic and orthographic variables of this study.

4.2 Variables and values

In the data, I annotated 2 response variables and 9 explanatory variables. The response variables are information status and information role – two information structural categories. The explanatory variables are various semantic, morphosyntactic and orthographic categories (see Table 3). Each variable’s values were annotated independently, separate from the values of other variables. It’s important to note that the values of the explanatory variables are not considered the defining or identifying criteria for the values of the response variables. The primary focus of this study lies in exploring the relationships between the response variables and the explanatory variables.

The values of the information structural response variables have been operationalized in terms of contextual relations. Specifically, the values of the response variables were annotated based on the informational relations between a linguistic unit and its surrounding context.

The response variable of information status has two values: information unit and information element. These two categories are distinguished by the involvement of a linguistic unit in at least one of the three information unit-internal relations: 1) the comment-forming relation, 2) the aboutness relation, or 3) the contextualizing relation between the frame and the rest of the information unit.

A linguistic unit directly linked to another linguistic unit through any of these relations is categorized as an information element. Conversely, a linguistic unit not directly linked to another linguistic unit through these relations is designated as an information unit. From another perspective, an information element has one or more information elements of the same information unit as its context, while an information unit has other information units as its context.

In terms of informational completeness, an information element expresses a part of an idea, whereas an information unit expresses a complete idea. In terms of internal information structure, an information unit possesses its own information structure, while an information element is a component of the information structure of a larger linguistic unit.

The response variable of information role concerns information elements. Information role has five values in this study: focus, background part of the comment, topic, frame and prominent element. The values of information role were annotated based on the way in which an information element relates to the other information elements within the information unit.

In this study, focus is understood as the most unpredictable information element in the information unit with respect to the context of the information unit. Such a definition combines Lambrecht’s (1994: 207) notion of pragmatic unpredictability with the notion of narrow focus (Valin and LaPolla 1997). According to this definition, every information unit has its own focus. The information structure of an information unit is pragmatically centered on the focus that is distinct from the foci of other information units. If an information unit consists of a single element, then this is the focus. In addition to the focus, an information unit may have other, unfocused information elements, that contextualize the focus and that are pragmatically backgrounded to the focus.

Usually, the focus can be questioned in relation to the other elements of the information unit. However, when dealing with subordinate clauses, determining the implicit question that the information unit addresses is not always straightforward. Since both the converb construction and its main clause can be subordinated clauses, I have not relied solely on questioning as the method for identifying the focus. Instead, I have relied on a number of interrelated and overlapping interpretative focus effects (see Matic and Wedgwood 2013). Specifically, I have identified the focus as the information element that most strongly evokes the interpretative effects of salience, remarkability, newsworthiness, and/or irreplaceability within the context of this information unit (see Matic and Wedgwood 2013: 137, 141, 157–158).

The unfocused information roles have been annotated based on their relationships within the information unit. The background part of the comment is an information element that provides immediate context to the focus. Combined with the focus, it offers information about the topic, whereas the topic itself is an information element that contextualizes the comment by indicating its subject matter. The frame is an information element that contextualizes the topic-comment structure, presenting an intermediary context that elaborates on the broader, information unit-external context of the information unit. The prominent element is any unfocused information element that is emphasized in relation to another information element of the same information unit. A summary of the characteristics of these information roles is provided in Table 1.

Table 2 gives an overview of the information status and information role categories used in this study.

In addition to the response variables, I have also annotated 9 explanatory variables (cf. Table 3) identified on the basis of previous studies (Lindström 2017; Remmel 1963; Simmul 2021; Tael 1988). Two of the explanatory variables – the number of words in the converb construction and the number of modifiers of the construction – are quantitative, while the rest are qualitative. Next, I will explain 3 explanatory variables that may not be self-evident.

As values of the morphological form (Form), I distinguish the three unmarked tense converbs of Estonian: the -des, -mata, and -maks forms. This variable makes it possible to find out how important the form of the converb is with respect to the information structure of the converb construction.

By the modifiers of the converb (NoM) I mean the direct dependents of converb: the object, the predicative and adjuncts of the converb. In (20) the converb has four modifiers.

By lexicogrammatical prominence markers (PrM)^[6] I mean non-constituent markers (21) and -gi/-ki clitics (22) that emphasize the converb or the construction. I also treat as a prominence marker the preposition ilma ‘without’, which emphasizes the absence of the event of the -mata construction.

Table 3 below gives an overview of all the explanatory variables and their values. The abbreviations used in the table are used in figures, tables, and examples throughout this article.

5.1 Information unit

A converb construction can function as an information unit, which expresses a complete idea, contains new information and relates to the main clause as its information unit-external context. It contains a focus of its own and may be preposed (28) or postposed (29) to the main clause or interposed between the parts of the main clause (30).

A preposed information unit presents a new idea, which in turn becomes the starting point of the idea of the main clause (31). A postposed information unit adds a new idea about the topic of the preceding main clause (32). An interposed information unit presents relevant information that stands relatively separate from the adjacent information structure of the main clause (33).

A typical converb construction functioning as an information unit contains a focus and a background part of the comment. The topic of the converb construction is implicitly understood from the context, usually from the main clause (examples 32–33). There are, however, instances (example 34) in which the converb construction arguably contains an explicit topic that reactivates the topic of the main clause. A word that often performs the role of reactivated topic is the adverbial pronoun ise ‘self’, which usually appears in constructions that contrast with the main clause.

A converb construction functioning as an information unit has an information structure of its own organized around a distinct focus. The converb construction expresses a complete idea and an event that is related to the event of the main clause, but is considered form a different perspective. Although syntactically the converb construction is embedded within the main clause, on the information level it stands apart from the other constituents of the main clause and relates to the rest of the main clause as a relative whole to a whole. The converb construction functioning as an information unit expresses sentential, eventive meaning and, on the information level, resembles a typical simple sentence, apart from usually not having an explicit topic of its own.

5.2 Element of an information unit

A converb construction functioning as an information element forms an information unit together with the main clause, wherein the converb construction has the role of frame, focus, background of the comment, or prominent element. In this data, none of the converb constructions function as a topic, i.e. there is no information unit whose comment is mainly about the event of the converb construction. This can be attributed to the fact that the dynamism of the event of the converb construction is not characteristic to the role of topic.^[7]

6.1 Variables relevant to information status

The random forest, developed by Breiman (2001), uses trial and error to determine the significance of each individual explanatory variable to the response variable (Baayen et al. 2013: 265, 267; Levshina 2015: 292; Tagliamonte and Baayen 2012: 157–158). The random forest measures how strongly each explanatory variable is correlated with the response variable. The interaction between the explanatory variables does not affect their relationship with the response variable. This is important in this study because some explanatory variables are closely related to each other. There is interaction between, for example, 1) the morphological form of the converb and the semantic function of the converb construction, 2) the position of the converb and the presence of punctuation, and 3) the number of modifiers of the converb and the number of words in the converb construction. Neither these nor other interactions increase or decrease the predictive power of any of the explanatory variables relative to the response variable. The random forest does not give any information about the relationships among the explanatory variables themselves; these relationships can be analyzed with the help of a classification tree (Baayen et al. 2013: 265, 273; cf. Figure 2).

Figure 1, based on a random forest analysis, arranges the explanatory variables from top to bottom in order of how strongly correlated they are to the information status of Estonian converb constructions. Seven relevant variables are shown by points located to the right of the vertical line on the diagram: the presence of punctuation (Punct; see Section 7.3), the position of the converb within the construction (PoCv), the semantic function of the construction (SF), the position of the construction (PoCC), the morphological form of the converb (Form), the number of words in the construction (NoW) and the number of modifiers of the converb (NoM).

Figure 1:

Variables relevant to the information status of the converb construction (random forest).

A classification tree gives an overview of the relationships of different explanatory variables to each other as well as to the response variable (Baayen et al. 2013: 265, 273). It identifies an explanatory variable closely tied to the response variable and splits observations into two sets on the basis of this explanatory variable (Baayen et al. 2013: 265; Levshina 2015: 291; Strobl et al. 2009: 325). This is then repeated in the subsets created in the previous iteration until either a) no more statistically significant variables can be found or b) the number of observations in any one subset reaches a given lower limit (Baayen et al. 2013: 265; Strobl et al. 2009: 327).

Although the principle behind the classification tree is simple, interpreting the diagram may be difficult if it contains a high number of nodes (Baayen et al. 2013: 265, 287). Thus, to optimize the explanatory power of a classification tree, a minimum number of observations required to form a subset is often established. In this study, I use a minimum of 7 % of observations, which means that each subset used in the analysis of information status contains at least 120 observations (cf. Figure 2) and each subset used in the analysis of information role contains at least 65 observations (cf. Section 6.3, Figure 4).

Figure 2:

Variables relevant to variation in information status (classification tree).

The classification tree (cf. Figure 2) predicts the variation in information status based on the presence of punctuation, the position of the converb and the semantic function of the construction. In the diagram showing the classification tree, divisions, i.e. nodes, are arranged from top to bottom: starting from the uppermost node (1), which divides all observations in the data set, each node (1, 2, 5) divides all observations that reach it into two groups. The groups (3, 4, 6, 7) at the bottom of the diagram show the summary totals of observations as bar charts, which show the proportions of different values of the response variable belonging to a particular group. The number of observations in each group is shown above the bar charts.

According to Figure 2, constructions separated by punctuation (node 1) typically function as information units (nodes 6 and 7). However, constructions that carry the functions concession, manner, means or time also function as information elements in almost one third of cases (node 7). Constructions not separated by punctuation usually function as information elements (nodes 3 and 4), but if the construction is converb-initial, it can also function as an information unit (node 4). In general, the presence of punctuation and the placement of the converb at the beginning of construction predict the status of information unit, while the absence of punctuation and the functions of concession, manner, means and time predict the status of information element.

6.2 Variables relevant to information role

Figure 3, based on a random forest analysis, arranges the explanatory variables from top to bottom in order of how strongly they correlate with information role. Three relevant variables emerge, shown by points located to the right of the vertical line on the diagram: the position of the converb construction (PoCC), the presence of a prominence marker (PrM) and the semantic function of the converb construction (SF).

Figure 3:

Variables relevant to information role (random forest).

The classification tree (cf. Figure 4) predicts the variation in information role based on the position and semantic function of the construction. The information role of converb constructions generally, i.e. except for the role of topic, follows the same word order patterns as the information role of a typical Estonian phrase (see Lindström 2017: 549–555). An information unit-final (postposed) converb construction typically acts as a focus (node 7), while an interposed construction typically functions as background of the comment (node 3). An information unit-initial converb construction functions as frame (node 5) or prominent element (node 6), depending mainly on its semantic function.

Figure 4:

Variables relevant to information role (classification tree).

The only coded variable that shows no significant relationship to the variation in either information status or information role is the presence of a deictic in the construction.

7.1 Morphological form of the converb

The distribution of information status and role of -des and -mata constructions differs from that of -maks constructions. Over half of -des and -mata constructions function as elements of an information unit, whereas almost two thirds of -maks constructions function as information units. Table 5 provides a breakdown of information status and role for the three morphological forms of the converb.

The differences in the information status and role of -des, -mata and -maks constructions are primarily explained by their differences in semantic function, word order and punctuation (see Sections 7.2, 7.3, 7.5 and 7.7). If semantic function, word order and punctuation are similar, then their information status and role are also likely to be similar, regardless of the morphological form of the converb. Due to the fact that there are explanatory variables that are more strongly correlated with the information status and role than the morphological form of the converb, the variation in the information structure of -des, -mata and -maks constructions can be predicted by the same general variables.

However, morphological form serves to explain some relatively regular cases in which some explanatory variable deviates from the general pattern. For example, morphological form helps to explain how the position of the -maks converb (see Section 7.2) and the punctuation of the -maks construction (see Section 7.3) relate to information status.

7.2 Position of the converb

99 % of Estonian converb constructions either begin or end with the converb.^[8] Both -des and -mata converbs may occur alone, as single-word constructions, and over half of -des and -mata converbs are the last constituents of the construction. -maks converbs, on the other hand, do not occur alone and are overwhelmingly positioned as the first constituents of the construction. Table 6 gives an overview of the distribution of converb position by the form of the converb.

Converb-initial constructions typically function as information units, while converb-final and single-word constructions usually function as elements of an information unit. Converb-initial constructions generally feature a construction-final focus, which is a characteristic of information units (48; see Section 3.2).

Converb-final and single-word constructions usually do not contain a focus; rather, they contextualize the focus of the main clause or function as the focus of the whole sentence (49).

Table 7 summarizes the relationship of the position of the converb with information status. Only construction-initial and construction-final converbs are shown in the table, because converbs positioned between modifiers are too rare (cf. Table 6) and single-word converb constructions always function as elements.

Roughly one third of converb-initial -maks constructions and one fifth of converb-final -mata constructions deviate from the general pattern described above. -maks constructions tend to begin with the converb regardless of information status (49).

The strong tendency of the -maks form to appear at the beginning of a construction could be explained by its artificial origin, stylistic markedness, association with specific registers, syntactic restrictedness, semantic meaning of purpose and its functions as a complement and a bound modifier (Erelt 2017c: 801–802; Uuspõld 1980: 729, 736). The relevance of these factors is a matter for a separate study, but in any case the -maks form tends to occur construction-initially regardless of information status.

Almost one fifth of converb-final -mata constructions function as information units, consisting solely of a focus (26) and complementing the preceding main clause with a supplementary piece of information.

In summary, the position of the converb indicates the presence of a focus, the construction’s informational completeness and information status. A construction beginning with the converb resembles a subject-less finite clause: it begins with the core element – the converb or predicate – and ends with a focused modifier. Single-word and converb-final constructions are similar to typical core-final (noun) phrases in Estonian, which usually do not contain a distinct focus and are not informationally independent.

7.3 Punctuation

Punctuation is the clearest formal indicator of the information status of Estonian converb constructions (cf. Figure 1). The relationship between punctuation and information structure is difficult to interpret, because in written Estonian, punctuation is regulated by rules and the material of this study consists of edited texts. However, punctuation is an empirical linguistic feature whose consideration as an explanatory variable does not diminish the value of other explanatory variables (see Section 6.2). Therefore, I will describe how strongly and in what way the punctuation of the converb construction is correlated with information status. The interpretation of the extent to which the relationship between punctuation and information status is based on the Simmul’s or editor’s perception of meaning, pause or rules is beyond the scope of the tree and forest method (see Section 6) as well as of this study.

The existence of punctuation rules does not in itself mean that punctuation is arbitrary in the sense of lack of information-structural function or motivation. Rules are one factor that affects the punctuation and that may interact in many ways with other factors, e.g. perception of meaning and pause. The punctuation rules concerning the Estonian converb constructions are not explicitly tied to information structure but nevertheless align well with information status and may be implicitly based on information structure.

Estonian punctuation rules have focused on word order, and mainly the position of the converb. According to the general rule, a converb-initial construction should be separated from the main clause by punctuation (example 23), but a converb-final or single-word construction should not be separated (example 24; Erelt 2006: 151–152; Saari 1993: 400; Vääri 1980: 144–145). On the level of information structure (see Section 7.2), this rule essentially implies that an information unit, i.e. a construction containing a focus, should be separated from the main clause, while an information element, i.e. a construction not containing a focus, should not be separated from the main clause. Thus there is a clear functional link between the punctuation rule and information status: punctuation marks the boundary between information units.

Most (96 %) Estonian converb constructions functioning as information units are separated from the main clause by punctuation (example 48), whereas most (95 %) of those functioning as elements are not (example 49). Sometimes, a converb construction functioning as an information unit occurs without punctuation (52). Most (72 %) of these cases are -maks constructions, which can be explained by the fact that punctuation in -maks constructions is by rule considered optional (Erelt 2006: 151–152) and the avoidance of punctuation has sometimes been preferred (see Saari 1993: 400).

On the basis of English, punctuation and intonation units have been treated as oral and written manifestations of information units (Chafe 1988; Halliday 1967: 201, 1985; Moore 2016). Similarly, it can be said that punctuation indicates the information status of Estonian converb constructions quite accurately, though not perfectly. The question through which processes the correspondence between punctuation and information status emerges requires a separate analysis.

7.4 Number of words in the converb construction and number of modifiers of the converb

A converb construction functioning as an information unit contains on average nearly twice as many words and one-third more modifiers than a construction functioning as an element. The relationship of information status to the number of words and modifiers is summarized in Table 8.

An information unit typically contains more words and modifiers than an information element. However, the relationship between information status and the number of words and modifiers is far from absolute. An information element (example 42) can be longer and contain more modifiers than an information unit (example 8). Thus the number of words and modifiers by itself is not a definitive indicator of information status.

7.5 Position of the converb construction

The position of the converb construction is to some extent correlated with its information status and strongly correlated with its information role. The distribution of position of constructions functioning as elements is relatively even, but constructions functioning as information units are usually postposed. Table 9 provides an overview of these relationships.

Converb constructions functioning as information units are rarely interposed, because a sentence in which one information unit is coopted (see Heine et al. 2017) between the parts of another requires tracking two ideas at once, which is complicated and risks violating the maxim of manner (see Grice 1975: 46).

The position of the converb construction is the main indicator of its information role. If a converb construction functions as an information element, its position relates to the information role like the position of a typical Estonian phrase, except for the role of topic (see Lindström 2017: 543, 549–550; Tael 1988: 38–40). A postposed construction usually functions as a focus (50), an interposed construction is typically in the role of background part of the comment (40), and a preposed construction is typically a frame (37). Table 10 gives a breakdown of the frequency of information roles in different positions.

The relationship of the position of the information element to information role is somewhat weakened by the prominent element role, as prominent elements occur relatively equally in both preposed (45) and interposed (46) positions. The characteristic emphasis of a prominent element is primarily marked not by position, but by the presence of a lexicogrammmatical marker.

7.6 Lexicogrammatical prominence markers

In Estonian, lexicogrammatical prominence markers are an important means of emphasis (Erelt 2017a: 53–54; Lindström 2017: 545–546). Such markers are found in roughly half (52 %) of converb constructions functioning as a prominent element (example 53) and in one ninth (11 %) of those functioning as a focus (54).

Converb constructions functioning as information units are rarely (4 %) emphasized by prominence markers. The roles of frame or background part of comment are not marked by prominence markers, because according to the definition used in this study (see Section 4.2), the emphasis of the prominence marker would change the role of frame and the role of background part of comment into the role of prominent element.

As the primary indicator of a focus in Estonian is its information unit-final position and an element in its typical word order position is usually not lexically marked (see Lindström 2017: 544–545), lexicogrammatical markers are used to highlight only those foci which are particularly notable, e.g. those that run counter to expectation (examples 43, 49, 54).

However, lexicogrammatical markers are the primary means of highlighting prominent elements, because the prominent – final and initial – word order positions are typically occupied by the focus and the topic or frame (see Lindström 2017: 547, 549, 552). The division of prominence-marking work between the focus and the prominent element is thus economical; the focus, as the more common information role, is typically marked by word order, while the less common role of prominent element is marked via additional lexicogrammatical means.

7.7 Semantic function

The relationship between semantic function and information structure is diverse, but some concordant patterns do emerge. Table 11 summarizes the occurrence of different semantic functions in different information status and roles. Functions 1–7 are characteristic of information units and functions 8–15 are characteristic of information elements.

The converb construction has a twofold meaning. It serves to expresses an event, while simultaneously indicating that this event is a circumstance of a larger event. In general it can be said that in the case of constructions with functions 1–7 the facet of an event tends to be contextually prominent. Conversely, for constructions with functions 8–15 the facet of a circumstance tends to be contextually prominent.

On the informational level, functions 1–7 tend to be associated with relative conceptual independence and completeness of the event. Converb constructions with functions 1–7 usually express a different situation (32) or the same situation from a different perspective (48) than the main clause. On the information level, converb constructions with these functions relate to the main clause as a comparative whole to a whole.

Functions 8–15, on the other hand, tend to be associated with informational partiality and conceptual incompleteness of the event of the converb construction. Expressing a circumstance of the event of the main clause, converb constructions with these functions relate to the main clause as a part to the whole. Furthermore, semantic functions 8–15 tend to be associated with different information roles. The role of focus is most characteristic of functions 8–10, in which the circumstance of the construction is interpreted as the most unpredictable aspect of the event of the sentence (example 53). The role of frame is most typical of functions 13–15, wherein the circumstance of the construction contextualizes the event of the main clause (example 36). Functions 11–12 are associated with the role of prominent element; constructions carrying this function tend to express a remarkable, though somewhat peripheral circumstance (example 52).

To some extent, the observed patterns of association between semantic function and information structure may apply to other constructions as well. This possibility is suggested by the typical morphosyntactic structures commonly associated with these functions in Estonian. Semantic functions specification, contrast, cause, result and purpose, characteristic of the information unit status, tend to be expressed by clausal-sentential and co-ordinated morphosyntactic structures, in contrast to functions like manner and means, which are more often expressed as phrases (see Erelt 2017b: 611–620; Erelt et al. 2020: 429–430; Hennoste 2017: 486; Veismann et al. 2017). The relationships between semantic function, information structure and morphosyntactic structure is a wide-ranging topic worthy of further study encompassing a variety of linguistic units.