(2.1) Gathering and organising the information

The first type of information displayed in the maps is the etymology. We mainly relied on de Vaan (2008) and Ernout and Meillet (2001), complementing them with Meiser (1998), Walde and Hofmann (1938–1956), and Meyer-Lübke (1935).

The core of the maps is the list of meanings (or functions), together with their first attestations, positioned with respect to a chronological axis. The description of the semantics of each modal marker is based on the ThLL, a monolingual monumental dictionary of Latin, written in Latin. The ThLL ideally reconstructs all the meanings attested for a lemma, listing the majority of the attestations of each meaning and organising them in chronological order. However, while the ThLL is the best and most complete dictionary of Latin (though the project is still ongoing), its description of modal markers is not compliant with any specific framework of a theory of modality. Therefore, building on the information given by the ThLL, we add our semantic analysis, which is oriented at identifying the modal meanings of each marker in a more fine-grained way. When relevant, we also add premodal and postmodal meanings, as defined in typological studies, in particular the one by van der Auwera and Plungian (1998). Sometimes the analysis of the attestations brings to light meanings, or, more often, specific nuances that were not mentioned by the ThLL. When this is the case, we include them in the description. For instance, one of the meanings illustrated by the ThLL for debeo is “to be obliged by necessity”. However, a closer analysis of the attestations revealed that this meaning could have three different modal readings depending on the context: dynamic necessity, deontic authority and deontic acceptability. We decided to represent in the map three instances of this meaning, one for each modal reading.⁶ At times, in particular in the case of collocations, we also needed to add one or more meanings to the ones already suggested by the ThLL. An example is the collocation necessitatem habeo in the map of necessitas ‘necessity’. The ThLL specifies all possible constructions (i.e., with the accusative necessitatem standalone or followed by a gerund(ive) in genitive or by an infinitive), but it does not associate it with any specific meaning. Building on the semantic analysis of the attestations, we assigned necessitatem habeo two meanings: ‘to show/imply a state of necessity’ and ‘to be under the necessity of’. The former is not modal, while the latter shows a basic value of necessity further specified into dynamic or deontic modality depending on the context.⁷

The changes or adjustments that we made to the semantic pieces of information given by the ThLL mainly resulted from a deeper analysis of the modal meanings of the relevant lemma. We rarely added or modified meanings that did not show any possible modal reading, unless it seemed necessary. As an example, we can mention the case of possum ‘to be strong, to be able’. For this verb, the ThLL presents a specific context of use, in which possum can be referred to words, money and similar. In this context, possum indicates the value of something: the economic value of money, or the semantic value of a word, i.e., its meaning. However, the ThLL does not provide a specific meaning description for these cases. Therefore, we decided to introduce the meaning ‘having some meaning or value’, as we did not want to discard this use of the verb, although not modal.⁸

The ThLL offers extremely detailed descriptions that are often organised in different sections depending on various criteria, such as the animacy of the participants, the context of use, or specific constructions of the lemma (e.g., with dative, with gerund(ive) in genitive or with the subordinating conjunction ut). As the maps need to be informative and legible at the same time, we partially discarded this type of information. However, we made some exceptions, based on our interest in the study of modality. For the modal marker licet ‘it is allowed’ the ThLL lists, among the others, the meaning ‘it is necessary’. This is in contrast with the basic value of possibility expressed by the marker, and with the other listed modal meanings (e.g., ‘it is allowed’, ‘it is permitted’, ‘to have the chance/possibility/opportunity of doing/undergoing something’). In fact, the ThLL adds that this meaning is attested in Plautus with an ironic nuance, where it is used as a mockery. In this case, the indication “ironically, hapax in Plautus” appears in the map next to the meaning ‘it is necessary’ in order to avoid possible confusion on the part of the user.⁹ As mentioned above, the maps also inform about syntactic structures or collocations in which the modal marker participates and that convey a specific meaning altogether. For instance, the lemma certus ‘certain’ can be used in collocations such as certum habeo ‘I consider it certain’, certo certius ‘certainly’, certior sum ‘to be informed’.¹⁰ Therefore, in the map of certus the user will also find these three collocations and their respective meanings.

In some cases, the ThLL organises the materials according to a semantic criterion, which we tried to preserve by gathering the meanings around the suggested groups. An example is the lemma incertus ‘uncertain’, whose description is divided into two sections: active sense, describing someone who is uncertain, and passive sense, relative to something doubtful, of which one should doubt. In the digital map, the two sections are represented as semantic groups, and each meaning is assigned to the relevant one. For instance, the meaning ‘lacking knowledge, unaware’ refers to the semantic group labelled ‘who is in doubt’, whereas ‘not determined, delimited, defined’ is assigned to the semantic group ‘doubtful’.

For some lemmas a description of the ThLL is not available yet. Therefore, we decided to rely on the Oxford Latin Dictionary—OLD (1968). This is the case for nolo ‘to want not’, (ne)queo ‘(not) to be able’, usus ‘use’ (with the modal meaning of ‘need’ arising in collocations such as usus venit and usus est), valeo ‘to have strength, to be able’, volo ‘to want’ and voluntas ‘will’. Although the OLD is undoubtedly less rich in information, especially about semantic groups and contexts, we applied the same procedure, by extracting the list of meanings with their first attestations and providing additional modal analysis. Specific constructions indicated by the dictionary were also listed as constructions/collocations, when pertinent.

Once all relevant pieces of information are gathered, the meanings are organised in space according to a chronological criterion. In this phase we provide an additional layer of analysis that aims at individuating the pathways of the semantic evolution of the lemma throughout time. This is done by detecting the relations between the different meanings of a lemma, based on their semantics and on previous studies. An important reference for this operation is represented by the models illustrated in van der Auwera and Plungian (1998), in a cross-linguistic perspective, and Magni (2005, 2010), for an application to Latin. The former elaborated a ground-breaking model of a cross-linguistic modal map, illustrating semantic relations involving premodal, modal and postmodal concepts in both synchrony and diachrony.¹¹ Building on this model, Magni (2005, 2010) developed a comprehensive map specific to Latin. Her map is based on the analyses of the modal pathways of debeo, licet, necesse (est) ‘(it is) necessary’, oportet ‘it is right, proper’, and possum (for a further discussion of the interplay between our maps and previous studies, see 3.2 below).

(2.2) From the draft to the digital version

The digital map is created by means of the tool Pygmalion. The information is encoded by filling out two forms. In the first form the user enters the lemma for which the map will be created and the etymological, chronological and semantic information, together with the attestations. The second form is dedicated to determining the relations between the meanings and their directions.

In the first form, the number of etymological steps can be set by the user. For the modal maps we encoded the Proto-Indo-European root and the Proto-Italic form that gave rise to the relevant Latin lemma. The user can also choose the width of the chronological subdivision to appear in the bar above the map: it can be set by centuries, decades or specific years. Our diachronic modal maps are based on centuries. Then, the meanings selected as pertinent are encoded one by one. For each meaning of the lemma we indicate: if it is conveyed by a collocation and which one (as for certus); if it is part of a larger semantic group (as for incertus); if it is modal (and what type of modality); the passage in which it is first attested; the century of the first attestation.

In the second form, the relations are drawn by selecting the two semantically related meanings and the direction of the link between them. It is also possible to leave the direction of the relation unspecified. We chose this last option for cases of two related meanings that are both first attested in the same century or, more generally speaking, when the directionality of the relation could not be established for some reason.

The etymological steps, the modal readings, or the relations between meanings can all be flagged as uncertain in the Pygmalion forms.

Once the two forms have been filled out, the map is automatically saved and can be downloaded as a JSON file. It can be visualised by uploading it on the Pygmalion platform at https://woposs.unine.ch/map.php.

(2.3) Supplementary material: location and format of the data

The supplementary material accompanying this article contains the data (i.e., the maps) and additional instructions on how to use them. The maps are stored in a repository in Zenodo.¹² The dataset contains: a README file which provides information about the contents of the dataset and the context in which the dataset was produced, and instructions on how to use the data; a JSON file for each modal map. The reason for using JSON format is that the tool employs the JavaScript library D3 for rendering the information in a visual form. Figure 2 shows how information about the meaning ‘ought, should (for logical or similar reasons)’ of the lemma debeo is encoded in a JSON file, and Figure 3 shows how it is rendered visually. The structure of a JSON file consists of pairs of name and corresponding value. In this case, the meaning ‘ought, should (for logical or similar reasons)’ is the value associated with the name ‘definition’. Constructions (including collocations) or semantic groups are specified by the two corresponding names, respectively ‘construct’ (an abbreviation for ‘construction’) and ‘group’. As in this case no constructions or semantic groups were associated with the meaning, their value is null, represented here with ‘_’. As more than one type of analysis is possible for a meaning, the information about semantic analysis is encoded in the array of objects ‘analysis’. Within this array, the name ‘category’ illustrates the modal (or not modal) reading (‘Modal: epistemic’), while the other names are associated with the date of emergence and disappearance of the meaning,¹³ the first attestation, the certainty of the modal analysis (defined by the boolean values ‘true’ or ‘false’) and the relationships between the concerned meaning and the other meanings in the map. The information about the relationships is stored in three different arrays depending on their type: ‘origins’ for directed relationships (ingoing arrow), ‘destinations’ for directed relationships (outgoing arrow), and ‘unspecified’ for undirected relationships (simple line). Each meaning (or semantic analysis, in case there are more modal readings for the same meaning) is assigned a random ID (in this case ‘mkny’). The IDs are then used as values to create the relations between meanings.

(3.1) Main results: An overview of an interactive map and its functionalities

In this section we will give an overview of an interactive map, taking possum as a reference example (Figure 4).

The etymology of the lemma is displayed in the arrow above the map, which is divided in as many sections as the number of etymological steps set by the user. In the case of possum, the arrow shows first the Indo-European root *pót-i-, followed by the Proto-Italic forms *poti- and *pot-ē- and their meanings, respectively ‘master, in control of’ and ‘to be master’, ending with the Latin paradigm possum, potuī¯, posse.

The chronological bar below, set by centuries, indicates the time span in which the attested meanings of the lemma appear.

All the meanings are displayed inside arrows. Each arrow begins in correspondence with the century in which the meaning was first attested. As the ThLL does not usually provide information about the date of (possible) disappearance of a meaning, the arrows continue until the end of the chronological bar. One exception is represented by the already mentioned meaning ‘it is necessary’ for licet (see 2.1 above). As the ThLL specifies that it is only attested in the Aulularia by Plautus, the arrow starts and ends in the 2nd-century BCE. The additional semantic analysis is color-coded: a legend at the beginning of the map explains the correspondences. For what concerns the modal readings, we have found attestations for the expression of dynamic possibility (dark blue), e.g., ‘to be able’; deontic authority (light blue), e.g., ‘it is allowed, fair, appropriate’; dynamic necessity (purple), ‘~ to have to’, where the symbol ‘~’ indicates an approximation of the use of possum in that context to the meaning ‘to have to’; and epistemic modality (green), ‘it is possible’. The meaning ‘it is possible’ with a deontic authority value is enclosed in dotted lines, which indicate that this type of reading is doubtful. Indeed, the deontic authority value seems to be attested only in this specific fragmentary passage. Since we do not have access to the context of this fragment, we cannot be sure about the modal reading. Non-modal meanings (grey) are ‘to have some meaning or value’ and ‘to be very powerful, to prevail’. We could not find any postmodal value for possum. On the other hand, the meaning ‘to be strong, powerful’ was analysed as premodal, leading to dynamic possibility values.¹⁴

One innovative feature of the maps is interactivity: the users can set the visualisation according to their needs. The first attestations are not displayed by default, as this would weigh down the visualisation. However, granting access to the first attestations is important for two reasons: each meaning is located on the chronological axis based on this information; moreover, the semantic context in which the marker appears is fundamental for determining a modal reading. In addition, in some cases different arrows can display meanings with the same wording but different modal readings. As shown in Figure 4, in the map of possum the meaning ‘it is possible’ can have a dynamic possibility, an epistemic or a (somewhat doubtful) deontic authority modal reading. The user can retrieve the different contexts by hovering over the arrow containing the selected meaning: the first attestation appears showing details about the author, the work and the line or verse corresponding to the attestation. We kept the ThLL abbreviations, so that the user can directly refer back to the index available on the ThLL website.¹⁵

Another interactive function is the visualisation of the semantic links between the meanings (Figure 5). If the directionality of the relation is specified, the link will appear as a line beginning or ending with an arrow. If the direction is not specified, the two meanings will be linked by a simple line. The user can display the relations by clicking on the arrow containing the selected meaning. This will show all the meanings that were at its origin (outcoming arrow) or were derived from it (incoming arrow). In order to visualise all the meanings again, the user just needs to double click on one of them.

The user can also choose the type of visualisation according to three sorting methods: chronological order, semantic groups and constructions/collocations. If the user selects the first option, the meanings are displayed from the most ancient to the most recent (Figure 4). The presence of semantic groups or constructions/collocations, if any, is not displayed with this option. The method ‘Groups’ shows the meanings gathered by a criterion defined by the user in the first Pygmalion form. The map of possum does not have any groups, but that of incertus, as mentioned in (2.1), shows two groups, labelled ‘doubtful’ and ‘someone who doubts’. The method ‘Constructions/Collocations’ offers a global visualisation of the meanings where those associated with a construction or collocation appear above (Figure 6).

Possum has two collocations: fieri potest ut ‘it can happen, be that’ (dynamic possibility) and potestur with passive infinitive ‘it is possible’, which receives two modal analyses (dynamic possibility and deontic authority).

The option ‘Constructions/Collocations’ is also used for forms of the same lemma associated with a specific meaning. We exploited this method to illustrate the Latin developments of Proto-Italic *lic-e-: alongside the impersonal modal verb licet, we find the active form liceo ‘to be for sale, salable’, ‘to have for sale’ or ‘to have the possibility/chance/opportunity of doing/undergoing something’ (the latter with a dynamic modal reading), and the middle liceor ‘to bid, make an offer’ or ‘to bargain, negotiate’.

Each lemma is also illustrated via a network graph shown below the map, where the nodes are the meanings and the edges are the semantic relations between the meanings (Figure 7). The edges can be directed or undirected, depending on whether we were able to establish the directionality of the relation. The network graph does not give information about attestations, semantic groups, or collocations. However, the chronological information is color-coded: the most ancient meanings are represented by darker blue nodes, and the shade of blue becomes lighter as the meanings become more recent. The network graph is meant to offer a synthetic overview of the meanings and their relations.

(3.2) Further results: the maps as a reference tool for studying modal markers

The digital modal maps were initially conceived for being used within the project in two main respects. First of all, they represent a support tool for the annotation task. They provide the annotator with a concise and intuitive overview of each modal marker to be annotated in the WoPoss corpus. At the same time, the maps are also used as a reference tool for our study of modal markers. At this stage, as explained in (2), they result from the organisation of lexicographic information in the theoretical and more abstract schema of the semantic maps of modality devised by van der Auwera and Plungian (1998) and Magni (2005, 2010). However, on a more advanced stage of the project, this first version of the maps will be compared against the evidence obtained from the annotated WoPoss corpus. The aim is to complement or modify the maps basing on the annotated data: this could concern a more ancient attestation for a modal marker or a modal meaning, a semantic nuance that was not captured on the base of the ThLL or the OLD, a new collocation, a change in the relations between the meanings.

Our maps also raise some interesting issues that we will illustrate by means of the case of licet. Magni’s work already showed that an attentive review of the Latin attestations can be revealing. For instance, she points out that possum is not a case of demodalisation as suggested by van der Auwera and Plungian (1998, p. 106), as the premodal meaning is attested early, already in Plautus (Magni, 2010, p. 217). Building on this, our maps contribute to the discussion in two main respects.

First of all, thanks to a clear chronological visualisation, they make evident when the possibility to establish the directions of diachronic changes cannot be based on the available attestations, but needs to be reconstructed on the basis of comprehensive and cross-linguistic studies. While semantic maps intended in the traditional sense offer a linear visualisation of the semantic changes, our maps, by integrating a philological and lexicographic approach, represent more clearly the actual situation we have to deal with, based on the available evidence. For instance, the map of licet shows a very early postmodal meaning conveyed by the construction ‘licet + subjunctive’, where licet has a concessive value. Notice that this meaning is as anciently attested as the deontic and dynamic ones.¹⁶

Our maps also allow us to highlight a caveat in the use of lexicographic sources for the study and representation of semantic changes, when these are not complemented with other studies. Although the ThLL is extremely rich in attestations, by relying entirely on this type of resource, we risk to end up neglecting rare (modal) meanings or specific semantic shifts. This is even more true for the OLD, which provides a smaller amount of attestations. The case of licet provides a good example. Magni, building on Núñez (1991, pp. 77–81), classifies licet as a modal verb as it can express both dynamic/deontic modality and epistemic modality. However, during our analysis of the attestations provided by lexicographic sources, we did not encounter any instance of epistemic modality for licet. The two authors present some evidence for this type of reading. The most striking example is given by Núñez (1991, p. 185):

1. Lucr. 5, 597 Illud item non est mirandum, qua ratione tantulus ille queat tantum sol mittere lumen, quod maria ac terras omnis caelumque rigando compleat et calido perfundat cuncta vapore. Nam licet hinc mundi patefactum totius unum largifluum fontem scatere atque erumpere lumen, ex omni mundo quia sic elementa vaporis undique conveniunt et sic coniectus eorum confluit, ex uno capite hic ut profluat ardor.

“Another thing also need not excite wonder, how it can be that so small a sun emits so much light, enough to fill with its flood seas and all lands and the heavens, and to suffuse all with warm heat. For it is possible that from this place is opened one single fountain of the whole world, to splash its generous flood and to fling forth light, because the elements of heat gather together from all parts of the world in such a manner, and their assemblage flows together in such a manner, that the heat flows out here from one single source.”¹⁷

In the text that follows this passage, not included for reasons of space, Lucretius offers two other hypotheses that could explain the phenomenon under discussion, i.e., the ability of the sun to emit so much light. Therefore, we can agree with Núñez on a predominant epistemic reading of licet in this passage. It is worth mentioning that the attestations offered by the ThLL did not include the passages cited by Magni and Núñez; moreover, both authors (Magni, 2010, p. 228; Núñez, 1991, p. 184) underline that epistemic readings of licet are rare. Building on this, we decided to add to the map of licet the epistemic reading, attested by this passage in Lucretius’ De rerum natura. This example proves that sometimes relying on lexicographic sources is not enough to provide a representation of the diachronic evolution of a modal marker as complete as possible. However, it also allows us to show the flexibility of our maps, which can be enriched with new information gathered from our sources, such as other researchers’ work (as in this case) or more extensive corpus data. In fact, a more comprehensive, corpus-based semantic analysis of the attestations could capture instances of relatively rare modal readings, as it was the case for licet, providing new and different insights on the evolution of modal markers in Latin. As already mentioned, we will be able to provide this type of analysis in the next stage of the project, by means of our annotated corpus.

(4.1) Implications for the practice

Thanks to the progress of computer science the field of data visualisation is rapidly evolving. Therefore, it is important to highlight possible implications of our visualisations in designing modality and modality-related issues, as well as their role as linguistic tools of synthesis and discovery.

Our maps represent the digital transposition of traditional non-interactive drawings which are found in many dictionaries and other types of publications. By exploiting computer technologies and in particular the tool Pygmalion, we combined this transposition not only with interactivity, but also with the possibility to add layers of information without overloading the design of the map.

From the point of view of more traditional linguistic approaches, our map model sits between that of lexicography—we will use the term ‘lexicographic maps’— and that of typology—their maps are usually called ‘semantic maps’ (for an application of semantic maps to the domain of modality, see van der Auwera & Plungian, 1998).¹⁸ It is beyond the scope of this paper to detail similarities and differences between diachronic lexicographic maps, traditional diachronic semantic maps and our diachronic maps of modality. Moreover, it must be underlined that the domain is rapidly evolving, as shown by the “lexical diachronic semantic maps” and the state of the art outlined by Georgakopoulos and Polis (2021). Notably, their model is based on network graphs and quantitative methods. Considering these recent developments, we can say that, while traditional lexicographic maps represent the etymological history of a lemma by focusing on formal and semantic changes and their relationships, traditional diachronic semantic maps—which often rely on cross-linguistic data—represent the organisation of semantic space with reference to meanings or functions and their relationships. Our diachronic maps of modality combine some of the features of the two traditional types of maps.

As diachronic lexicographic maps and diachronic semantic maps, our modal maps present the diachronic feature too, but they offer a more precise anchoring to the temporal dimension thanks to the chronological bar. This represents a drastic difference in particular with typological maps: the latter are usually meant to be universally valid generalisations, and therefore are not tied to a specific temporal dimension. Moreover, our maps contain much more data about the represented items, as illustrated in (2) and (3).

Another similarity shared by all the aforementioned types of maps is the representation of the relationships between the meanings or functions (or more generally the represented items). However, in our maps, thanks to their interactivity, the relationships between the meanings only appear when clicking on an arrow. This way, the visualisation is not weighed down by an excessive amount of information, and the user can select specific items and focus on the relations of each one of them individually.

To conclude, while lexicographic and semantic maps are intended to offer synthetic overviews, our maps are rather conceived as tools to interactively explore various types of information. In fact, the major contribution of our map model to the practice of designing maps is probably the fact of allowing the user to explore a big amount of data at the same time.

(4.2) Applications

Our maps have a wide range of applications that go beyond the scope of the WoPoss project itself.

They could complement the available lexicographic resources for these lemmas, providing a finer analysis of their modal readings. As shown above, e.g., for possum, this type of analysis is not present in the ThLL nor the OLD, even when they mention the basic modal meaning(s) for a lemma.

We also find that the maps can represent a benchmark for scholars interested in different topics, such as modality and evolution of modal pathways, semantic change, history of the Latin language, Latin linguistics. This is especially due to the fact that they summarise in an interactive image a considerable amount of information that is gathered from many different sources. The consistent description provided for each lemma can orient scholars in their research, saving them a time-consuming task, selecting the most relevant information and providing additional analysis on modality that cannot be found in any of these sources. If necessary, the files of our maps can be downloaded and changed according to the requirements of various circumstances.

Another field of application for the maps could be the classroom. The benefits of involving visualisation techniques in the teaching and learning process are known in literature (Arcavi, 2003; Jessop, 2008; Rattya, 2013; Klerkx, Verbert & Duval, 2014 among others). The existing maps can be used to teach about modal categories and subcategories and their use in context, but also about more general (historical, semantic and partially syntactic, via the collocations) aspects of the Latin language. For instance, they could be used to introduce pupils and students to the use of impersonal verbs. Our set of modal maps features some impersonal verbs or constructions, e.g., respectively decet, licet, oportet and aequus est, necesse est, meum est among others. Thanks to the option ‘Constructions/Collocations’ and to the attestations associated with each meaning, the learners can access the syntactic aspects of these verbs and constructions. Moreover, the interactivity of the maps makes them suitable tools for individual learning: the levels and/or types of information presented in the map (etymology, semantics, chronology, constructions and collocations) are well separated, and learners can easily adapt the visualisation to their learning stage and objectives.