Finding a model for contrastive lexical semantics: A look at verbal communication verbs

A basic problem for contrastive lexical studies in general is to find a model for the semantic analysis. This paper is one in a series of corpus-based contrastive studies of the field of Verbal Communication Verbs (VCVs) in English and Swedish. Searle’s classification of speech acts serves as an important starting point but is not directly concerned with lexical structure, which is a major concern for the two theories that are compared in this study. FrameNet based on Fillmore’s theory of semantic frames and Wierzbicka’s theory of semantic primitives (or “primes”). The theories are applied and tested on data from the English Swedish Parallel Corpus (ESPC) containing English and Swedish original texts together with their translations into the other language. Primarily two groups of English verbs and their Swedish correspondents will be analyzed: (1) Information verbs such as tell, inform, notify, report, narrate and describe and (2) Speech activity verbs such as talk, speak, chat, converse, gossip, discuss, debate, negotiate and bargain. There is also an analysis of Swedish berätta ‘tell, narrate’ based on the Multilingual Parallel Corpus (MPC) as an example of multilingual contrastive analysis. Frames relate in a clear way the conceptual structure and the syntactic argument structure, which is very useful in a contrastive study. However, the definition of the meaning of individual verbs is incomplete and needs to be complemented with some kind of decompositional analysis such as the theory of semantic primes. A special section is devoted to an analysis of a large number of compound and derived forms of the Swedish verb tala ‘speak’ and a discussion of how contrasts in morphological structure can affect the lexical contrasts between two languages.

patterns of polysemy that characterize them. The two earlier studies present rather detailed descriptions of the most frequent VCVs. Viberg (2017a) was concerned with the English verbs say, tell, speak and talk and their Swedish correspondents and the other study (Viberg, 2016) with verbs of asking and answering in both languages. Even if a few verbs dominate the field, the total number of VCVs is very large. Ballmer and Brennenstuhl (1981) present a taxonomy of no less than 4,800 English speech act verbs (including multiword expressions). The present study will discuss various ways to approach the description of more fine-grained distinctions between VCVs in English and Swedish. The major aim is to identify important concepts for the analysis rather than to give complete analyses of individual verbs.
A basic problem is to find a model that accounts for both semantic contrasts and contrasts in the grammatical realization of semantic distinctions. With that aim in mind, various frameworks will be compared. 1 In particular, the analysis of semantic frames in FrameNet will be complemented with elements from componential approaches such as Wierzbicka (1987) and Miller & Johnson-Laird (1976). The bulk of the paper shows how the analysis can be applied to various groups of VCVs based on corpus data. A special section is devoted to a discussion of how contrasts in morphological structure can affect the lexical contrasts between two languages.

Verbal Communication Verbs (VCVs). A Brief Overview
The major types of verbal communication verbs are shown together with a few illustrative examples in Table 1. The verb say holds a unique position with respect to both its high frequency and its semantic generality and is referred to as the nuclear verb of the field (for nuclear verbs, see Viberg, 1993). Verbs foreground (or profile) various aspects of the communication frame. Descriptive communication verbs profile other aspects of the communication than the message. The most important descriptive verbs, in particular speak and talk, refer to the Speech Activity as such without specifying what is said (the message). In the unmarked case, the nuclear verb say as well as speak and talk refer to oral communication, but these verbs can be used to refer to communication in other media (The letter says…, She talks about this in her mail). There are a number of verbs that profile one specific medium, most important of which is write ranked as the fourth most frequent verbal communication verb in both English and Swedish. Verbs like murmur, stutter and shout refer to the Manner of speaking (such as acoustic and rhythmical properties, emotion and intelligibility). The largest and most important group is represented by the Speech Act verbs, which are used to perform and report speech acts in the sense of Austin (1962). Table 1 follows Searle's (1976) classification of speech acts with some modifications, although the present paper will primarily deal with reported speech acts. Primarily two groups of English verbs and their Swedish correspondents will be analyzed.
(1) Information verbs such as tell, inform, notify, report, narrate, describe, explain and lecture; (2) Speech activity verbs such as talk, speak, chat, converse, gossip, discuss, debate, negotiate and bargain. In addition, the role of the morphological structure will be discussed based on the derivational possibilities of the Swedish verb tala 'speak'.

Corpora
The analysis will be applied and tested on data from the English Swedish Parallel Corpus (ESPC) consisting of English and Swedish original texts together with their translations into the other language (Altenberg & Aijmer, 2000). The English original texts contain around 705,000 words and the Swedish around 661,000 words. The ESPC is very well-organized. Fiction can be compared to Non-Fiction and the corpus can also be used as a comparable corpus by comparing the original texts in both languages. Below examples from the ESPC will be shown followed by a text code within parentheses based on the author's name (for a list of texts and codes, see Electronic sources). Data will also be taken from the Multilingual Parallel Corpus (MPC), which at present consists of extracts from 22 Swedish novels and their translations into English, German, French and Finnish (around 600,000 words in the Swedish originals). Examples from this corpus will be marked MPC followed by a text code (see Viberg, 2013 for a list of the texts and codes).

Frames in FrameNet
Frame semantics is the theoretical foundation of FrameNet, a large computerized lexical database for English (see Fillmore, Johnson & Petruck, 2003 and Electronic sources). In this theory, verbs (and other relational words) evoke frames -schematic structures of recurring situations (Fillmore, 1982). A frame consists of various frame elements (FEs) such as the ones for Communication shown in Table 2. Verbal communication is a more specific frame that inherits the structure from the Communication frame, which includes both verbal and non-verbal communication. The FE Communicator is represented by Speaker, which is a more specific FE. There are many frames that inherit their structure from Verbal communication, among them Telling, which is defined as follows: "A Speaker addresses an Addressee with a Message, which may be indirectly referred to as a Topic. Instead of (or in addition to) a Speaker, a Medium may also be mentioned.  v, apprise.v, assurance.n, assure.v, brief.v, confide.v, inform.v, let know.v, notification.n, notify.v, tell.v. (FrameNet also analyzes nouns, but in this paper only verbs will be discussed.) Table 3 shows some examples of the syntactic realization of the frame Telling. Only the phrase structure is indicated. It is possible to add one more layer showing grammatical relations such as Subject and Object, which will also be referred to in the analysis. In language teaching, one of the advantages with FrameNet is that it can be used to show, for example, the choice of prepositions marking the Topic. On the other hand, there is no systematic account of the semantic differentiation between verbs that belong to the same frame, such as tell, confide and brief. There is also a Swedish FrameNet (see Electronic sources), which presents a similar analysis of the Telling frame in Swedish. Wierzbicka (1987) in her semantic dictionary of English speech act verbs manages to distinguish more than 200 VCVs in a systematic way by using "reductionary paraphrases" based on a restricted set of primitives (or "primes"). Tell, for example, is represented as in Table 4. The definition of each verb is accompanied by an elegant essay, which it is hard to do justice to in a short space. The wording of the definition is intended to show that tell has a more general meaning than just to impart knowledge. Among other things, the verb is used to refer to the telling of fictional stories that the Addressee wants to hear but not necessarily uses to acquire factual knowledge. Wierzbicka (1987) divides the VCVs into 37 groups. The TELL group contains tell1, report1,2, narrate, relate, recount, describe, explain and lecture1,2. (Some verbs refer to more than one speech act and belong to more than one group. Tell2, 'tell sb to do sth', belongs to the ORDER group). A definition is provided for each verb in a group, which makes it possible to distinguish each member semantically. The definition of tell can be compared to the definition of narrate in Table 5. The definition highlights among other things that narrate requires many events and emphasizes the temporal order of events. The definitions of other verbs in the group make it possible to make fine-grained distinctions. Relate purports to refer to real events and requires a fairly complex subject matter but does not like narrate require a narrative line but can rather refer to a flow of situations. Recount like narrate emphasizes the sequential nature of the events but is closer to relate by paying more attention to descriptive detail. The set of semantic primes in Wierzbicka (1987) has been further developed together with Cliff Goddard into the Natural Semantic Metalanguage (NSM, see e.g. Goddard & Wierzbicka, 1994;2014). For VCVs, the prime "say" is central and appears in the definitions of all the other VCVs. The primes of NSM cannot be further analyzed. However, it is an open question how far the decomposition should go. Miller & Johnson-Laird (1976) regard their predicate SAY as a combination of producing sounds (The parrot said 'goodbye') and conveying meaning (The signpost said 'Ten miles to London') (Symbolized as UTTER(x, W) and MEAN(x, v).

Information Verbs
Speech acts that are used to describe the world are referred to as Representatives by Searle (1976) and can be reported with two types of verbs: Assertive verbs, which report what is regarded as an opinion by the Reporter, and Information verbs, which report information that the Reporter more or less accepts as a fact. This section will present a corpus-based contrastive analysis of a number of information verbs. Such verbs include the verbs invoking the Telling frame in Frame Net and the verbs belonging to Wierzbicka's TELL group and the closely related INFORM group: inform, inform on, notify, announce and reveal. It will not be possible to analyze all of these verbs. The selection is restricted to the most frequent verbs and to some related verbs that are reasonably well represented in the two translation corpora.

Tell and its closest correspondents
Tell and its closest correspondents in Swedish are analyzed based on data from the ESPC in Viberg (2017a). The major translations of tell (as an information verb) are shown in Table 6. The most frequent translation is säga 'say', which supports the claim that tell has a rather general meaning as an information verb. There are also two frequent verbs that are more direct correspondents in the sense that they both have tell as their dominant translation, when Swedish original texts are translated into English. (English say is used as a translation of these two verbs only in a few cases.) As is further elaborated in Viberg (2017a), berätta tends to refer to a relatively complex message, whereas tala 'om tends to relate an individual fact. Intuitively, the prototypical meaning of berätta is centered around berätta en historia 'to tell a story', even if the majority of the actual examples refer to a less complex message. Clearly the meaning is not as general as that of tell (see Table 4 above). Actually, Wierzbicka's definition of narrate (see Table 5) comes close to the prototypical meaning of berätta. However, narrate is an infrequent, rather technical term and does not occur even once as a translation of berätta in the ESPC. As a complement to Viberg (2017a), this study will present data from the MPC showing how the two verbs berätta and tala 'om in original Swedish texts are translated into English, German, French and Finnish. Example (1) shows the verbs that most frequently are used as translations of berätta and the realization of the FEs Message and Addressee.
In Swedish, the Addressee is realized as a PP with the preposition för, which typically marks an Expriencer, whereas it is realized as an object in English and is marked as a dative in German and French (and à + Noun). In Finnish, it is marked with the allative case (-lle), which signals the Receiver but also motion to a place. The Message is realized as an indirect report (a that-S clause in Swedish, English and French, whereas German uses a specific subjunctive clause and Finnish a participial construction that can replace a that-S clause.) Data about berätta are summed up in Table 7. The extent to which various FEs are explicitly realized contrasts across languages. The section at the top of Table 7 shows that the Message is usually (in 84% of the cases) realized explicitly in Swedish, whereas the Addressee is realized only in 26% of the cases. This means that berätta focuses on the Message. In English original texts (Viberg, 2017a), tell has an explicit Addressee in 90% of the cases. The simple exchange in (2)-(3) shows how this plays out in context. In (2b), only the Addressee is explicit, but the Message is implicitly retrievable from (2a) and the same applies to (2c). In (3b), which is the Swedish original text (with tala om, the other major correspondent of tell), the Addressee is explicit, since it represents new information that is not implicit in the context, but in (3c), where both the Addressee and the Message are mentioned in the preceding context, the Message is explicit (det 'it') in Swedish but not the Addressee (as in English 2c).
a. "Jag flyttar ut i skogen nu", sa Ronja. There are several parallel examples that illustrate that tell focuses on the Addressee to a greater extent than berätta, whereas the Swedish verb to a greater extent focuses on the Messsage. As can be observed in Table 7, one verb dominates as a translation of berätta in all languages. However, the degree to which that applies varies greatly. Finnish kertoa is used as a translation as much as in 92% of the cases, whereas French raconter only reaches 51%. (German erzählen: 81%, English tell: 74%). One reason why French raconter is used less than the major translations in the other languages has to do with the use of a Topic as in (4) The Topic marked with a preposition in all languages except Finnish where it has a case marker, the elative (-sta in Karmanjaka-sta), which basically has a spatial meaning ('out of'). What is particularly noteworthy is that berätta in (4) is not translated with raconter in French but with parler 'speak/talk' that is a speech activity verb. As can be observed in Table 7, where Speech activity verbs are placed on the same line as talk, parler is used as a translation more frequently than its correspondents in the other languages. Parler is primarily used when berätta has an explicit Topic (49 examples out of the 59, when parler is used as translation). Actually, raconter is also used as a translation 43 times (out of totally 175), when berätta has an explicit Topic, but in those cases the Swedish Topic often corresponds to an object of raconter as in (5)  When berätta is combined with an explicit Topic it functions as an atelic speech activity verb rather than a telic speech act verb. This can be tested with durational adverbs: Peter berättade om sin hund i två timmar 'Peter (?told me)/ talked about his dog for two hours' vs. Peter berättade historien på fem minuter 'Peter told (me) the story in five minutes'. It appears that raconter stays closer to the prototype ('tell a story') than Swedish berätta does. In order to test this, French original texts must be included in the analysis. The fact that tell is used as a translation of berätta as often as in 74% of the cases seems to indicate a close semantic relationship, but, as was demonstrated above, translations from English into Swedish show that tell has a much more general meaning than berätta. Actually, in addition to the Swedish original texts analyzed in Table 7, the MPC also contains extracts from six French original novels and their Swedish translations. The French original texts contain 23 raconter, out of which 19 are translated by berätta. The Swedish translations of the same texts contain 60 berätta. In addition to the 19 raconter, berätta is used as a translation of parler 15 times. Even if data are limited, they clearly indicate that raconter does not (like English tell) have a large set of uses that are not covered by berätta. The extension of raconter obviously is more restricted than that of berätta. Table 8 shows the most frequent translations of tala 'om. It turns out that this verb has the same dominant translation as berätta, except in French where dire 'say' is more frequent as a translation than raconter. This means that the contrast between berätta and tala 'om is language specific and has no direct semantic equivalent in the other languages. As mentioned, berätta has the prototype 'to tell a story' and in general tends to refer to a complex set of events or facts, whereas tala 'om tends to refer to a single fact. English, German and Finnish do not make this distinction and use one verb to cover both meanings, whereas raconter stays closer to the prototype of berätta. For that reason, the major option is to use the most general speech act verb dire 'say' to translate tala 'om. The analysis presented above only gives examples of some important contrasts but should suffice to show how a close study of the distribution and realization of frame elements across languages can be used to explain the variable patterns of translational correspondence.

More Specific Information Verbs
Both English and Swedish have a relatively large number of more specific information verbs. This section will give some examples how these are used to express more specific meanings and how these verbs are related to the most basic information verbs.
Focus on the Addressee. Information verbs are concerned both with the specific factual information contained in the message and who has been given that information. In face to face interaction, production and reception of the message occurs simultaneously (for practical purposes). In institutional settings, a message such as a decision by the board can be drawn up by a small group of people and then disseminated later to a wider group of people. There are a number of verbs that typically are associated with such situations even if some such verbs also are used more generally in informal conversations. Verbs like inform and notify in English and the Swedish verbs informera, upplysa, meddela and underrätta are examples of this type of verbs and focus on the Addressee. All these verbs take the Addresse as a canonical object that can be passivized. In (6) and (7), the original Message is already known to the receiver of the reported speech.
There had to be people I should notify, and I didn't know who. (DF1) Det måste finnas personer som jag borde meddela, och jag visste inte vilka. Table 9, the Addressee is expressed as an object in the unmarked case. The Topic is also relatively often expressed explicitly as in (8), but the Message more seldom.

As shown in
The Council shall inform the European Parliament of its recommendation. (MAAS1) Rådet skall underrätta Europaparlamentet om sin rekommendation.
When the Message is expressed, the Swedish verbs tend to keep the Topic preposition om 'about' as in (9). She informed us that she planned to spend that night, (JSM1) Hon upplyste oss om att hon tänkte stanna över natten, [She informed us about that-S] Focus on the Topic. Verbs like English describe and characterize and their Swedish correspondents beskriva och karakterisera focus on the Topic which is realized as an object (see 10). The Addressee is optional and realized as a PP.

He described his religion to the king, (LT1) Han beskrev sin religion för kungen
The Message can be realized as a description marked with as/som and indicates the category of the Topic in the form of an NP (see 11) or indicates its properties with one or more Adjective (He is often described as irresolute). The Topic is often passivized as in (12). Jag tror det kan beskrivas som ett slags förälskelse.(PE1) I believe it could be described as some kind of falling in love. Describe and the verbs of similar type cannot take a that-S complement but can take a how-S or what-S complement (see 13). Such complements are Topics rather than Messages.
He described how he proved the superiority of Norwegian ski over the North American snow shoe. (RH1) Han beskrev hur han hade bevisat de norska skidornas överlägsenhet över de nordamerikanska snöskorna, Both in English and Swedish, there are a number of more specific verbs that pattern like describe (see 14) and are based on the metaphor that something is described in pictures or painted in words (WORDS ARE PICTURES), for example English depict and portray and Swedish utmåla (lit. 'out'-+ 'paint') and skildra (etymologically derived from German schilderen 'paint a coat of arms', cf. Germ. Schild 'shield').
Yet to modern ears, this is a horrible story: it depicts God as a despotic and capricious sadist (KAR1) /---/ Den utmålar Gud som en despotisk och nyckfull sadist,  Dirven et al. (1982) and Viberg (2017a) compares these two verbs with their closest correspondents in Swedish tala, prata and snacka. After a brief summary of an earlier study of speak and talk directly below, the following section will show how the analysis can be extended to the non-basic, more specialized speech activity verbs.

Speaking and Talking
Speak and talk focus on the linguistic activity as such, not the Message, as say does (see Viberg, 2017a). The distinctive feature of talk is that it focuses on linguistic action as discourse. The subject of talk often has the role as speaker-interactor and the verb is also topic-prone and can take the topic as object in a way that speak does not (talk shop/politics etc.). Speak, on the other hand, can focus on the actual speech production (Don't speak Che Guevara's name) and can have the code as direct object (speak English). In Swedish, there is no parallel to the semantic contrasts between talk and speak. The contrast between prata, snacka and tala (as a speech activity verb) is rather stylistic but also depends on the formality of the speech situation to which the verb refers. Tala tends to refer to a formal speech situation and is the verb that is used to refer to the delivery of a speech. In addition, tala has a rich potential to form phrasal verbs and compound and derived verbs with a specialized meaning, which will be discussed in the last section before the Conclusion.

Specialized Speech Activity Verbs
Speak and in particular talk have large sets of hyponyms/troponyms, for example chat, converse, gossip, discuss, debate, negotiate and bargain in English and småprata, samtala, konversera, skvallra, debattera and diskutera in Swedish. All of these focus on interaction and contrast by referring to specific types of topics and goals of the talking. Chat and småprata ('small' + 'talk') refer to informal social talk for relaxation and topics that should not be too serious (see 15).
Konverseraconverse also refers to social talk but topics can be a little more serious. In Swedish, the Addressee can appear in the object slot as in (16).
The Swedish verb samtala ('together' + 'talk') is more frequent than småprata and konversera. Like these two words it focuses on interaction (see 17) but otherwise it is more general and is used as a correspondent of converse and talk and even of speak together and chat.
Gossip and skvallra refer to indiscreet talk about other people's weaknesses (see 18) but need not be very serious. Förtala is a derivation of tala and refers to malicious gossip and legally refers to defamation.
He said, "I don't gossip with Harold, Ginny." (JSM1) Han sa: "Jag skvallrar inte med Harold, Ginny." Diskuteradiscuss and debatteradebate focus on verbal interaction but also have a focus on the topic which appears in the object slot as in (19). See also Table 11.
The topic is a real object that can be passivized (see 20) (cf. the object of talk: *Business was talked).
Olika åsikter debatterades öppet i en flora av tidskrifter och i tidningar (AA) Various opinions were now debated freely in a flora of journals and news sheet The aim of a debate is to present arguments for and against certain ideas and requires that there are at least two opposite views. The aim of a discussion is rather to reach a decision. That is explicitly signaled when the topic is realized as an if-S or WH-S complement in the object slot (see 21).
At that age you don't discuss whether or not you're going to be friends, you just are. (JB1) I den åldern diskuterar man inte om man ska vara vänner eller ej -man bara är det.
In academic texts, discuss is often used in a weaker sense as in (22) to indicate that a specific topic is treated more in detail in a specific part of the text. Typically, the verb is in passive form. This kind of realization, where the Topic appears in the subject slot, is shown in Table 12.
The imaginary exercise of seeding Mars with life, or even of bringing Mars to life, is discussed in chapter 8. (JL1) Den tänkta uppgiften att beså Mars med liv, eller till och med att ge Mars liv, diskuteras i kapitel 8.
In Swedish, behandla 'treat', 'deal with' is rather often used as an alternative translation (see 23).
Many matters discussed in this annual report (PHA1) Många av de frågor som behandlas i denna årsredovisning [that are treated]

Tala and the Morphological Potential of Swedish
Contrasts between languages in the general morphological structure contribute to lexical differences. Swedish has a greater potential than English to form derived verbs that are transparently related to the most basic verbs within a semantic field. There is a continuum between grammatical markers that function as free words and bound grammatical markers (cf. the use of case to mark Addressee and Topic in Finnish). For that reason it is motivated to begin by looking at prepositions, which are important in both English and Swedish as markers of Addressee and Topic. In Swedish med 'with' is the dominant marker of the Addressee with tala and practically the only marker when prata and snacka are used. This preposition marks the Addressee as an interactor and underscores the interactive nature of prata and snacka. In English, to is the unmarked preposition and is often used even when the Addressee is conceptualized as an interactor. With is used only exceptionally to mark the Addressee as in (24). In Swedish, till 'to' can be used only with tala and unambiguously signals that the talking is mono-directional.
Han talade till George, men han talade med Augusta. (SCO1) He was talking to George, but with Augusta.
Worth mentioning is also a special case where English and Swedish are similar. Both languages mark what Dirven et al (1982) refers to as a transmitter with the corresponding prepositions throughgenom (see 25). A transmitter (e.g. a spokesperson) only produces the speech which expresses the message of the original sender which is treated grammatically as the Speaker.
Sedan talar han till mig genom Maria. (PE1) Then he talks to me through Maria.
Talking to oneself without another person as addressee is signaled with the preposition för 'for' in Swedish (see 26) and can be used with all three speech activity verbs referring to overt speech (cf. säga sigsay to oneself, which can refer to silent talk, i.e. to thinking).
These are the ones most likely to talk to themselves. (MA1) Det är de som oftast pratar för sig själva.
The preposition för, which basically marks an Experiencer or a Beneficiary, appears in a number of other combinations that are worth noting. It can be used when speaking on behalf or in favour of someone else, for examples somebody who is simply speaking on behalf of the tobacco industry (EJAC1) translated: någon som helt enkelt talar för tobaksindustrin (cf. Tala för dig själv -Speak for yourself (don't take for granted that I agree). In Swedish, för can be used as a marker of a plausible conclusion, whereas the subject indicates the kind of support on which the conclusion is based ( Lexical contrasts must be judged against general structural differences between the languages being compared. There are important differences between English and Swedish with respect to word formation (compounding and derivation). Comparing speech activity verbs in English and German, René Dirven says: "To put the case in a somewhat extreme way, German thrives on its morphological potential to create forms for meanings; English, in contrast, has to create all sorts of syntactic devices and tap the metaphorical potential of language to cope with its conceptual drive" (Dirven, 1990, p. 261). Swedish similar to German has a rich morphological potential. The derivational possibilities of tala are demonstrated in Table 13. For comparison, parallel structures with verb particles and prepositions are shown in the last two lines. The derivational processes are general and very productive. They are exploited to a very great extent to form VCVs. Only a small proportion of the Swedish VCVs are simple verbs. The number of Romance borrowings is also low in comparison to English. Note: Stress is marked with an apostrophe, which does not appear in the standard orthography.
Compounding is very productive even if many of the compounds have low frequency and several of them are occasional. A typical example is valtala 'make an election speech' consisting of the noun val 'election' and the verb tala. It is possible to add another noun: EU-valtala 'make an election speech as a candidate for the EU parliament'. An important characteristic of Swedish compounds is that they are pronounced with word accent II. A selection of additional examples is given in Table 14 to give an idea of the productivity. Most of these compounds refer to formal or institutionalized speech events and the nouns in most cases indicate the type of event.  , inledningstala, öppningstala, invigningstala, avslutningstala, premiärtala, jungfrutala, Almedalstala, hyllningstala, middagstala, avskedstala, högtidstala, griftetala, festtala, majtala, förstamajtala, sommartala, gästtala, brandtala, buktala Swedish has a number of prefixes, which by definition cannot occur as free words, i.e. they are non-separable. There are two very frequent prefixes, which are unstressed: förand be-. Both are used with tala. Förtala, which means means 'slander', is an old verb but still is transparently related to tala, whereas betala, which means 'pay' and is a borrowing from Middle Low German betalen (be-+ talen 'count'), probably is not experienced as related to tala by most speakers. (However, the noun tal still has the meaning 'number' in addition to 'speech'). There are also derivations with a stressed non-separable prefix such as samtala 'converse, discuss' and åtala 'prosecute'. These verbs are pronounced with accent II and form a transition to the large and important group that will be treated next. Many combinations of verb + particle can be used both with the particle as a free word and with the particle preposed as the first element in a compound verb with accent II. The use with the particle as a free word is unmarked and the most frequent alternative. It can be used both when the meaning is literal and when it is figurative. As demonstrated in Johannisson (1954), bound forms usually have a nonliteral meaning. Occasionally, a bound form appears with literal meaning but then it tends to be very formal or sound obsolescent. Compare examples such as (28)  When tala is used with the free particle in as in tala in (ett meddelande), tala has its literal meaning, whereas the meaning is abstract when the particle is bound as in intala (sig, reflexive) 'persuade (oneself)' based on spatial metaphors (IDEAS ARE THINGS and THE MIND IS A CONTAINER: 'to put an idea into a person's mind by talking').
Several prepositions can also be used as particles. Then they are stressed and follow special word order rules. This applies to om, which is unstressed and used as a preposition in 'tala om 'talk about' and stressed and used as a verbal particle in tala 'om 'tell' (see above). There is also an initial bound form: omtala 'mention'. Sometimes it is not obvious which form is more abstract as in uttala 'pronounce' and tala ut 'speak one's mind', but this does not contradict the general rule that free particles can be used with a literal or figurative meaning, whereas the bound form in general is figurative. The variety of derivations and alternations between free and bound forms of many prepositions and verbal particles represents a rich potential to express a large number of separate meanings which often are relatively transparently related, at the same time as there is a high degree of conventionalization and specialization of the meaning. The derived forms of tala are shown in Table 15. The semantically and stylistically regulated alternation between compound and free forms in Swedish is different from the corresponding alternation in German, which is one of the phenomena that has attracted most attention in German word formation. In German, the alternation between prefixed and free particles in separable verbs such as aussprechen (out-+ speak) 'pronounce', zusagen (to-+ say) and einreden (in-+ talk) 'persuade' is grammatically regulated, whereas there are no grammatical restrictions on verbs in Swedish except that participles always have bound particles. For example, in German the bound form must be used in the infinitive (see example 29 above). In addition, the infinitive marker zu appears between the bound particle ein-'in-' and the verb stem reden 'talk'. On the other hand, free forms must be used in the present and past tense in main clauses, compare Ich rede mich ein -Jag intalar mig 'I persuade myself'. In Swedish, there is a change of meaning, if the free form is used: Jag talar in ett meddelande 'I am recording a message'. Subordinate clauses only allow bound forms in German but subordination does not affect the choice between bound and free forms in Swedish. In Swedish, it is possible to exploit the formal alternation between free and bound forms to signal semantic contrasts and to create new lexical items. Often the bound forms have a very specific meaning and are used in specialized registers (administrative and legal language, academic prose). In some cases, the alternation between free and bound forms is only stylistic without any clear semantic differentiation, but that applies to rather few combinations (e.g. upphöra and höra upp can both be used in the sense 'cease, stop').
In an ongoing study based on the ESPC, all compound verbs containing the bound forms of the spatial particles in/ut and upp/ner ('down') are studied. In all there are 37 such compound verbs (types) that can be used as verbal communication verbs in the ESPC (see Viberg, 2017b Section 4.3.2).

Conclusion
This paper has compared FrameNet and Wierzbicka's theory of semantic primitives as models for corpus-based contrastive lexical studies. Wierzbicka's (1987) semantic dictionary of English speech act verbs presents detailed descriptions of more than 200 VCVs and can be used as a model for a similar analysis of other languages. The theory of semantic primes aims at giving unique definitions of each individual verb and accounts for fine-grained distinctions between verbs that are closely related semantically. On the other hand, there is no clear link to the syntactic realization of the argument structure.
FrameNet provides a rather robust account of the basic conceptual structure of sentences with VCVs in terms of the core frame elements (FEs) and also accounts for the grammatical realization of the argument structure of the verb. An additional advantage of FrameNet is that it covers a rather extensive part of the lexicon in English and a growing number of other languages, among them Swedish (see Electronic sources). A major limitation of FrameNet is that it does not provide a unique semantic (or conceptual) representation of each lexical unit that belongs to a certain frame.
In a corpus-based contrastive study based on FrameNet, it is possible to study contrasts between correspondents by looking quantitatively at different patterns of realization of FEs as was done for tell and berätta. Tell and berätta are described with the same FEs but still differ in meaning. One way to describe this contrast is to say that tell focuses on (or profiles) the Addressee, whereas berätta focuses on the Message. However, some reference to semantic primes or some other componential theory is needed as well, for example, to account for the language-specific contrast between berätta and tala om in Swedish.
A closer look at the way FEs are realized with words belonging to the same frame as was done, for example, with inform above, makes it possible to give a more fine-grained analysis of words with related meaning. In this way, contrastive comparison can contribute to the further development of framenets of individual languages.
Most frameworks for contrastive lexical analysis seem to ignore the role of differences in the general morphological structure between the languages that are compared. This is highlighted in the last section above, but much remains to be done to integrate morphology into a general model of semantically based lexical comparison.