We did not consider proper interjections or filled pauses: they perform functions that are similar to those fulfilled by DMs and may have a procedural meaning, but they have no other non-discursive function from which the discursive function has developed (Borreguero, 2015). Moreover, we also excluded several frequent constructs of contemporary

Italian speech, such as quello che, lit. ‘that which is’, considered by Cella (2015: 160) to be an example of structural change currently underway. The structure is formed by the demonstrative pronoun quello, ‘the one’ — which either functions as the (direct or indirect) object of a preceding verb or is linked to a noun — leading to a relative clause introduced by che, ‘that’, followed by the verb to be (acting as the copula) and a noun. Let us consider the example in (1) from which we could remove quello che è without losing anything from a semantic point of view:

The construct shares several functions with those performed by DMs, such as gaining precious time to search for the most suitable word to say or mitigating (see D’Achille, 2016), but, once again, according to the criteria adopted in this work, it does not present defining properties of DMs because it does not gain new discourse functions or values.

Similarly, let us consider the following example (2) extracted from a webpage (Sansò, 2020: 48):

Sansò explains that the entire proposition non lo pensi anche tu…!? (‘don’t you think…!?’) could be considered as a DM, because it performs the same discourse functions (a request for confirmation/agreement) as no?, ‘isn’t it?’ or vero? ‘right?’. In a scale of explicitness and transparency, it is situated in the maximum positive pole. Moreover, the author claims that it is preferable to adopt a broad view with the procedural meaning as the only discriminating element, not least because only in this way can the different diachronic paths of these elements be examined. Indeed, fully transparent procedural expressions often turn into opaque elements (as in the case of (I) dunno, coming from I don’t know what else to say about this topic, Sansò, 2020: 50).

Nevertheless, we do not agree with this view: from a synchronic perspective, the procedural meaning of this kind of expression (such as non lo pensi anche tu…!?) does not derive from a desemantization process but can be ascribed to the more general metalinguistic function whereby, when we speak, we continuously give instructions for interpretation to our interlocutors.

For the same reasons, if we compare two expressions with a topic-orientation function (resumption) such as tornando a quanto dicevo prima (lit. ‘going back to what I was saying before’, ‘back to my point’) and in ogni caso (lit. ‘in any case’, ‘anyway’), only the latter was considered to be a DM in the present study, assuming it does not have the conventional meaning of “in each of the cases considered” but serves to resume a previously mentioned topic. Both can perform the same function in discourse but, in our restrictive view, only in ogni caso (lit. ‘in any case’) can be considered a DM with multifunctionality and a primarily pragmatic meaning, while examples like the former rely on their conventional meaning and “are more appropriately thought of as conjunct adverbials (if phrasal, e.g. back to my point) or as clause chunks (if they have a clausal structure, e.g. I totally forgot)” as discussed by Traugott (2020: 19) regarding the marker back to my point.

The dataset is derived from a corpus of Italian tourist guides’ speech, the CHROME Corpus, collected within the CHROME project (Cultural Heritage Resources Orienting Multimodal Experience, Origlia et al., 2018; Alfano et al., 2023). The CHROME corpus consists of audio-visual recordings of three female Neapolitan tourist guides, each leading four tours at the San Martino Charterhouse in Naples. Each tour is organized into six architectural and artistic Points of Interest (PoIs):

In each PoI one camera was aimed at the guide while another was aimed at the audience, and speech was recorded by three microphones: one on a headset worn by the TG and two at about one meter equidistant from the TG and directed towards the tourists. Thus, even though the tour took place in both mobile and stationary interactional spaces, speech recordings were made in stationary spaces, corresponding to the six PoIs. Although tourists could freely intervene by interrupting the TG’s explanations, they did not do so often.

Before moving on to the next PoI, TG asked if there were any questions. Hence, Q&A generally occurred at the end of the explanation of a PoI and technical considerations mean that speech in mobile interactional spaces does not form part of the recordings. This should be kept in mind when the study is placed within the overall framework of mobile vs. stationary interactional spaces of the interlocutors while talking and walking (Mondada, 2009, 2014a, 2017, 2018; Broth & Mondada, 2013). Our focus is exclusively on the TGs’ speech rather than on the audience’s, although the latter’s presence is crucial to determine this complex expert-tonon-experts interaction (Calvi, 2005), with an inevitably varying degree of audience specialization.

TGs have the important task of providing an adequate key to interpreting historical, artistic and natural heritage, while achieving mediation between specialized language and everyday speech, typical of communication between experts and the general public.

Table 1 shows the CHROME dataset analyzed in this study, corresponding to 7 PoIs for each TG. The oral texts have an overall duration of 3 hours 21minutes 37 seconds.

DMs were identified by following the criteria laid out in Section 2. The annotation was performed three times by one coder (the author) over a period of two years and disagreements between the three versions helped to improve the annotation scheme and the consistency of annotations. Criteria that were not sufficiently operational or replicable were removed or reformulated, particularly where fuzzy boundaries between labels were concerned.

The identification procedure and the label assignment were performed by carefully assessing the context and evaluating the procedural meaning(s) of the DM. In a second step, the same labels were compared in order to check that they annotated the same function(s). Although we will focus on examples in microsequences, the analysis is based on the whole conversation they are extrapolated from.

Labelling was conducted using the ELAN software for multilevel linguistic annotations (Sloetjes & Wittenburg, 2008), where the category labels for each tier are assigned from a pre-specified vocabulary. Each tier in ELAN corresponds to a dimension or a macro-function (M-F) with its functions indicated by mutually exclusive labels entered in the ELAN vocabulary. In the type of speech considered in this study, two tags may co-occur only in the interactional dimension: a request for confirmation may also be turn-yielding, but in our corpus this is very rare. However, when it did occur, we did not consider the tag indicating turn management, assuming it to be implicit and labeled it as a request for confirmation. On the contrary, there are no a priori constraints between the different annotation dimensions. Therefore, every DM can receive several labels in the different dimensions so that the typical polyfunctionality of these elements can be respected.

On a separate level, speech was segmented into major prosodic units, identified by clear acoustic boundaries (Degand & Simon, 2009), in order to consider the position of the DM. In this work, we do not address the relation between functions and positions of DMs, for which we refer to Alfano & Schettino (2023). Major prosodic units were isolated by considering a number of phonetic boundary markers, namely the presence of a (potential) final pause; f₀ declination of both f₀ and energy; parametrical reset at the beginning of a new TU; prepausal lengthening. We segmented speech by following perceptive criteria and, when in doubt, we moved from ELAN to Praat (Boersma & Weenink, 2019) and manually followed the rules for splitting speech into major prosodic units identified by Degand & Simon (2009). These units, often delimited by real or potential pauses, are generally characterized by an initial f₀ reset, followed by a declination, an overall fall in f₀ and intensity, and a variety of final contours. In our examples, we place them between square brackets, using a transcription system based on Savy (2005). Table 2 shows the main symbols used for orthographic transcription.

As for the translation of the Italian DMs into English, we indicate a literal translation for the item under investigation, which corresponds to one of the first senses of the lemma in the Collins Unabridged Italian to English online dictionary. This choice aims at a simple reproduction of the form in order to give an idea of the conventional meaning associated with it. In many cases, the translation of a DM is not only particularly difficult (Aijmer et al., 2006; Bazzanella, 2006a), but the best translation may simply be its omission in the target language or its replacement by another lexical item (Borreguero, 2011). If the examples contain other occurrences of DMs, in addition to the form under investigation, the proposed translation seeks to comply with the procedural meaning, values and functions played by the DM.

In addition, a separate ELAN tier was used to indicate clusters of DMs, identifying co-occurring DMs with the same function (catene, ‘chains’, Bazzanella, 2011):

or with different functions (cumuli, ‘clusters’, Bazzanella, 2011):

However, unlike Bazzanella (2011), we classified as catene and cumuli only consecutive co-occurring DMs in the same major prosodic unit (not simply “juxtaposed”, see Cuenca & Crible, 2019), exemplified in (5) and (6) respectively. In the name of the audio file, G indicates the guide, V the tour and P the point of interest. Henceforth, we indicate it in brackets for each example:

In (5) and (6) we find clusters of DMs, but in (5) both cioè and ovvero perform the same reformulation function with explicative purposes (Flores Acuña, 2009; Dal Negro & Fiorentini, 2014; Solsona, 2014; Sansò & Fiorentini, 2017): they are used to illustrate, expand or clarify the previous segment (“procurators were quite important figures and were functional to life at the Charterhouse”). On the contrary, in (6) quindi, assolutamente and diciamo perform different functions. Quindi is used to indicate a deduction of a logical consequence of something said or inferable: the crucifixion of Christ is presented as a quintessential classical element, which justifies the chosen epithet. Assolutamente functions as a boosting device, to increase intensity relating to the propositional content and illocutionary force of the deduction. Finally, diciamo fulfills the function of speech management, covering the programming times necessary for speech production (the cooccurrence with the filled pause <eh> reinforces this kind of interpretation, see Waltereit, 2006) while at the same time responding to a preparatory tempering mitigation strategy, and functions as a bush to reduce the precision of the propositional content (Caffi, 2007). The lexical choice classicheggiante, not a classic style but a style that imitates or tends toward the classical, has an approximation value in itself, reinforced by the DM diciamo, which functions as a further approximating device. Clearly, functions are assigned on a contextual basis: each occurrence of diciamo does not necessarily perform both functions and, obviously, may perform others (reformulation, for instance).

The tag-set follows the model devised by Bazzanella (1995, 2006a, 2006b, 2011) and other scholars (López & Borreguero, 2010; Molinelli, 2014; Borreguero, 2015; Sansò, 2020) and includes three macro-functions (M-Fs): interactional, metatextual and cognitive M-Fs.

The interactional M-F comprises the functions that serve to manage the interaction and social identity of speakers (for instance, those operations that pertain to turn management or phatic communication). The metatextual M-F consists of those functions related to the organization and the structuring of the discourse as well as the linguistic formulation of the text (e.g., the introduction, change, and closing of topics and the strategies to plan an utterance). Lastly, the cognitive M-F encompasses the functions related to cognitive processes, such as inference, speaker’s subjectivity and commitment, propositional content and illocutionary force.

Our tag-set is not intended to be exhaustive, but encompasses the functions identified in our corpus, summarized at the end of Section 3 in Table 3. Although the examples contain DMs that perform more than one function simultaneously, below we only illustrate the one for the label we wish to exemplify. Remember that square brackets indicate major prosodic units. Where there are exchanges between the TG and the audience, the latter’s turns are quoted in a new line and indicated explicitly by “audience:”. If there is no indication, it is the TG’s speech.

The following sections will offer an example for each function and will explain the functions that appear to be less shared, in order to clarify which criteria were used in assigning the respective labels.

Our analysis focuses on 1759 occurrences of DMs. If we look at the incidence of DMs per 100 words, we obtain a general frequency close to 6 % (5.78 %, or, to employ a more usual normalized frequency in Corpus Linguistics, a frequency of 57.80 per thousand words (ptw)). In order to understand the actual incidence of these elements, we have considered orthographic words, excluding nonverbal vocal elements (indicated in hooked brackets in the transcription, such as filled pauses). Therefore, in a string made up of repetitions or truncated words, each one counted as a word: sulla nel soff+ sul soffitto (lit. ‘on the in the ceil+ on the ceiling’) counted as 5 words.

If we look at M-Fs and their combinations marked by a “+” (Figure 1), not surprisingly the prevailing dimension is the metatextual one, both in isolation and in combination with the cognitive M-F.

The description of the M-Fs provides a general overview, but the analysis of the various functions within each macro-type offers a more accurate picture. In the interactional dimension (see Figure 2), we can see a distribution of various functions, but there is a clear prevalence of the function Marking of shared knowledge (M-S-K). DMs that function as turn-taking (T-T) and politeness (Pol) devices are obviously present. Finally, there are DMs that function as a clear attempt at direct audience involvement: Request for attention (R-At) and Request for agreement and/or confirmation (R-A-C).

Moving to the metatextual dimension (see Figure 3), we again observe a strong prevalence of one type: Plan. In almost half the cases, DMs serve to manage speech flow, gaining precious seconds for planning and production. The next most frequent functions are Paraphrasing and Focusing, with paraphrastic reformulation strategies (Paraph, Ex) outnumbering the non-paraphrastic ones (Cor, Reform). While Cor and Reform are more clearly non-paraphrastic, it is debatable whether or not Ex is. We emphasize this finding to note the prevalence of an explanatory type of reformulation, which highlights the TG’s need to provide explanations. However, besides the semantic information, the reformulated segment may also convey a change in the argumentative orientation, in the presuppositions and implicatures and in the polyphonic structure (Pons, 2013).

As far as topic management is concerned, we see that DMs are used more to shift to another topic (Shift), to close it (Clo), to go on with the same topic (Cont) or to make a reference to a previously mentioned not surprisingly, (Ref) than to introduce (Intro) or resume (Res) it.

In the cognitive dimension (see Figure 4), we observe the presence of all types of DMs considered, but Inference (Inf) and Mitigation (Mit) DMs prevail over Intensification (Int).

Having provided an overview of the functions performed, we now focus on the forms they take. Table 4 shows the frequency list of the first 20 DMs found in the dataset (a total of 1759 tokens, 102 types) and the number of functions performed by each form. They account for 84 % of the total, while “other” refers to a total of 82 forms that occur less than 13 times. The first 20 DMs are not only very frequent, but also very polyfunctional: the number of functions performed by each form averages 5.65, indicating that relatively few forms may perform many different functions.

The distribution of these more frequent forms indicates that the first three DMs, quindi, appunto and diciamo account for 49 % of the total (see Table 4). On the other hand, if we look at less frequent DMs (“other” in Table 4), 90 % of these forms perform only one function.

Low semantic definition, which does not mean low information content or low interpretability, is described as the most functional strategy in face-to-face communication, since it allows for incremental progress in meaning construction (Voghera, 2017). This strategy is implemented in speech through the use of highfunctional performance words or expressions with a wide range of meanings, i.e. those that show greater adaptability and flexibility in a greater number of contexts. Despite certain obvious differences, this preference also seems to manifest itself for DMs.

Their polyfunctionality makes them more employable for more purposes and in more contexts and thus encourages their greater frequency of use (see Degand et al., 2022).

As for syntagmatic polyfunctionality, that is to say how many DMs perform more than one function at the same time, we identify a value of 19 % in our dataset, with a marked prevalence of the association between metatextual and cognitive functions (17 %), while the other two combinations (metatextual M-F + interactional M-F and interactional M-F + cognitive M-F, only account for 1 % (see Figure 1). The most frequent functional associations are shown in Table 5.

Plan + Mit is the prevalent expression of syntagmatic polyfunctionality in our dataset: in this case the DM helps to manage the discourse flow by preventing occurrences of disfluencies and, at the same time, functions as an approximating device that reduces the precision and the adequacy of the lexical choice: for instance, diciamo in example (6) is simultaneously a speech management and an approximation device.

Inf DMs also function simultaneously as topic management. In (30), we see an instance of a Clo DM:

Here the TG had presented the altar and emphasized the fact that, despite appearances, it is made of wood, namely a poor material. She then moves on to the angels on either side of the altar and concludes that, again, it is made of papier mâché, a poor material. In the context, the defining feature of the material used is whether or not it is a poor material. Hence, the deduction made explicit by the DM quindi also indicates a topic closure: we can conclude that all elements of the altar are made of poor materials.

Finally, Inf DMs also function simultaneously as Paraph DMs, as in the following example (31):

Here a relationship of semantic identity is established between the reference segment and the reformulated segment, in the sense that the first Carthusian monastery had been built in the year 1300. At the same time, equivalence is also indicated by the DM as inferable information, made explicit by the reformulated segment.

As far as co-occurrences are concerned, consecutive co-occurring DMs account for 12 % of the total; 6 % of the DMs appear in cumuli (clusters of DMs, each one with different functions) and 6 % occur in catene (clusters of DMs, all performing the same function or functions).

Cumuli do not present restrictions with respect to the combination of functions. Indeed, their distribution does not show statistically significant trends (x²= 81.77, df = 54, p = 0.069): each function may co-occur with other ones. However, combining both monofunctional and polyfunctional DMs, the functions that most often occur in cumuli are monofunctional DMs of Plan (12 %), Inf (10 %) and M-S-K (8 %), that is, they can be found in all three M-Fs. As regards the sequencing of monofunctional DMs, there are no rigid orders among the functions, but looking at the M-Fs, we can see that, when present, interactional DMs tend to appear first, then metatextual DMs, and then cognitive DMs. Figure 5 depicts functions in cumuli, without considering functions that have a 1 % occurrence rate (which collectively account for 9 % of the total).

Figure 5 points out that DMs occurring in cumuli are mainly monofunctional (85 %). Indeed, syntagmatic polyfunctionality in cumuli can take the form of several combinations, but none are very frequent.

Catene show a narrower range of functions than cumuli (see Figure 6) and present clearer trends: the functions most likely to appear in consecutive co-occurrences are Focusing and Planning, reaching statistical significance (x²= 206.74, df = 54, p = 0.0005).

Like cumuli, DMs occurring in catene are mainly monofunctional (80 %). In this kind of speech, the most frequent polyfunctional DMs in catene serve to plan and mitigate. Both cumuli and catene almost always consist of sequences of two elements and rarely of three DMs (see Alfano & Schettino, 2023).