Linguistic and semantic layers for emergency plans

Abstract

Plans for emergency response are complex collaborations in which actors take roles and responsibilities. They are generally long textual documents containing practical instructions, in natural language, for hazard responses. A more rigorous structured-text would be useful for a twofold audience. From one side, it can be useful for quickly understanding the plan and on the other side it can be used to improve the modelling phase and delivering an automatic emergency-support system. This paper proposes an approach, conceived for humans, for converting a free-form plan document into a structured version of the same document. The approach is based on a linguistic and semantic analysis that are strictly correlated and materialize in a metamodel. It contains the essential elements of an emergency plan, and it aids in interpreting the input document also reducing inconsistencies, redundancies, and ambiguities.

Keywords

Emergency plans metamodel free-form to structured text

1 Introduction

Worldwide, emergency response plans describe complex processes involving collaboration and interaction of multiple roles and departments. In Italy, the definition of emergency plans is ruled by a normative layer that provides a national standard for managing emergencies. These rules originated by the Emperor Octavian Augustus, who firmly believed that “the value of planning diminishes with the complexity of the state of affairs”. In the first century before Christ, he put the basis of modern planning strategies that are now collected into the “Augustus Method” [18], introduced in 1997. This document proposes a uniform planning approach that is, at the same time, simple and flexible, where the key idea is to overcome the classical approach based on the bureaucratic census of equipment used in civil protection interventions with a new focus on assets’ availability.

In the context of the project N.E.T.TUN.IT 1 , we are working on a fully operational platform for cross-border data collaboration to cope with shared risks and disasters due to emergency scenarios. The project objective is that the Italian and Tunisian sides collaborate in the response to a simulated accident impacting nearby population, causing health risks as well as atmospheric and marine pollution.

Emergency response plans are typically expressed as informal documents written in natural language. This represents an obstacle for applying a systematic modelling, as well as for automatic verification and run-time adaptation. Nonetheless, understanding a free-text document (e.g. with the purpose of extracting fundamental information) is a challenging task that requires skills and domain knowledge.

In [6], we started a preliminary work for identifying a number of keywords in the free-text version of the paper, collecting and sorting their specific meanings. These words are so recurrent in the text that suggested us to be representative for describing organizational patterns during emergency response activities. The work concluded that some keywords in the text could be used as a pivot for extracting the intricate meaning of an emergency plan.

Now, building on those bases, this paper addresses the problem in a more systematic way by establishing a semantic layer for emergency plans. We work under two basic assumptions: i) we limit the scope of the study to Italian plans, and ii) the study specifically focuses on organizational patterns that emerge from plans. The objective of the work is to provide a set of guidelines for manually converting a free-form text into a structured one, in which semantics is made explicit. The present work presents three novelties with respect to the previous paper [6], as listed below:

The first novelty is a clear separation of the linguistic and semantic aspects through the analysis of lexicon, with the aim of reducing the gap between the lexicon and the metamodel.

The second contribution is the metamodel for modelling an organization for emergency response. Differently, by other approaches in the literature [3, 45], our proposed metamodel originates from the empirical study of existing emergency plans, and it is well grounded in the linguistic analysis of the text.

The third contribution is a list of linguistic issues to be aware when reading emergency plan documents. Even if, in this paper, we only reason about their possible presence in the input text, their categorization represents a first step towards a subsequent work of delivering a (semi-)automatic transformation from free-text to structured-text documents.

The rest of the paper is structured as follows: Section 2 provides an overview of the long term agenda on which the current work relies, and it provides a motivating example for justifying the importance of the proposed approach. Section 3 introduces the Italian landscape in the definition of emergency plans and illustrates the state-of-the-art about ontology and metamodels for emergency management. Section 4 presents the result of a linguistic analysis of the selected corpus of documents we used to empirically define our semantics, that is materialized in a metamodel, presented in Subsection 4.2, actually the core of the paper. Section 5 offers a few examples for providing a gradual understanding of how to use the metamodel (several others are reported in Appendix C). Section 6 reports our experimental setup, discusses the results, and it also provides a discussion of the limits of the current approach. Section 7 provides an overview of the whole project, describes future directions and possible challenges, whereas some conclusions are drafted in Section 8. Finally, three appendixes report in details property tables for the elements of the metamodel, sentence templates for rewriting in a structured way the input text, and several examples.

2 Motivation: from free text to structured plans

The objective of this section is to provide motivations for the current work. To this purpose, we illustrate, by means of an example, the need for a semantic layer when dealing with emergency management plans. Along the whole paper, we report many examples of sentences taken from real emergency plans. It is worth noting that, despite the original sentences being in Italian, we report translations, as accurately as possible.

An emergency plan is usually provided in free-text form according to a generic format prescribed by national laws. As an instance, we report below an excerpt from a plan [13, p. 93,94]siracusa[10, p. 55]conceived to deal with a fire emergency also involving the civil protection agency, volunteer organizations, and the police:

[...] The head of civil protection agency activates its cartography support service at the civil protection headquarters to prepare useful maps for the management of the emergency and the subsequent phase of return to normality. [...]

This kind of documents may present flaws due to the natural language such as redundancy, ambiguity, and the use of synonyms; moreover, different authors may express the guidelines by using their own sensibility and linguistic skills thus introducing an undesired alteration of the intended plan. As an example, one of the issues is the use of different words (synonym) to address the same event or the same action. It is common, when writing, the voluntary use of synonyms for the sake of elegance and for avoiding word repetitions. Even if that leads to more elegant documents for humans, it carries uncertainties that may cause ambiguities in the interpretation of the plan, and it represents an obstacle for an automatic processing of the document.

To this purpose, we highlight the need to convert informal documents to more rigorous models in which semantics is highlighted and uncertainty is removed. We suggest a two-steps approach, in which the first point is resolving linguistic issues and the second one is identifying essential elements to reveal the document structure. The first challenge is to construct a linguistic layer for analysing an input emergency plan text, with the aim of dealing with the uncertainty of the natural language and resolving linguistic flaws.

The linguist analysis, will be connected to a semantic layer, intended to analyse the structure of sentences and show in small and compact semantic networks the concepts hidden across the document. Here, the challenge is focusing on the organizational structures that are recurrent in many emergency management plans. The use of a metamodel facilitates the reader in generating a well-structured version of such documents.

The expected result is a semantic network like the one shown in Figure 1, built starting from the previous exemplar sentence. The underlying semantic of the nodes and relations of this diagram will be detailed later on in this paper. Here, it is worth noting how the same sentence is structured by assigning precise meaning to the various components of the discourse.

Fig. 1

Instance of the metamodel for the exemplar sentence reported in Section 2.

Despite the existence of automatic tools for interpreting the natural language, it is worth highlighting the role of manual translations of free-text in the context of emergency plans. This is a requirement that has its roots in the particular nature of the domain and in specific requirements of the project.

A recent study, mainly focusing on Chinese emergency plans, proposes automatic extraction of business process models from textual descriptions [20]. Despite the great advantage in terms of time and resources, the authors recognize a lower quality with respect to manually processed texts.

Interestingly, semantic role labelling systems [25, 35] proved to perform reasonably well in some controlled experiments. However, a degradation of performance has been demonstrated when a supervised system is faced with unseen events or a when the testing corpus is different from the training one [25]. Therefore, applying semantic role labelling to the crisis management plan domain could represent a big challenge, surely out of the scope of the N.E.T.TUN.IT. project.

In addition, manual translation is motivated by the central role of the humans in crisis management. This occurs at different levels of the process, starting from the responsibility to set and approve the emergency plans, and up to the need of maintaining a complete control of the ongoing activities [7]. In this context, a totally automated approach will break the precise distribution of responsibilities, mandatory by laws. Clearly, this point does not exclude that future semi-automatic approaches could facilitate this translation.

Another interesting point to clarify is how the resulting semantic networks (like the one in Figure 1) could be usefully employed in an emergency management. In our vision, the structured version of the input corpus of documents will support further phases of conceptual modelling and, consequently, an execution layer. Just to provide an intuition of this phase, let us consider the previous emergency plan as an example of transformation from structured-text to modelling notation. For the sake of simplicity, we adopt the very well-known BPMN notation to represent, in a visual format, the same corpus of information (see Fig. 2).

Fig. 2

A fragment of BPMN derived from the exemplar sentence.

Finally, Section 7 provides an intuition of the long-term agenda, related to the N.E.T.TUN.IT project, in which the final objective is to exploit a frameworkfor the adaptive execution and management of emergency plans.

3 State of the art

In this section, we first provide a bird-eye view of the Italian landscape about emergency management, with a specific focus on national norms and regulations. In the second part, we review some existing approaches to highlight the semantic of an emergency plan.

3.1 The italian landscape

In Italy, the structure of emergency response plans follows the so-called “Augustus Method” [18]. This method was developed by the Italian Civil Protection Agency and officially adopted by the Italian Ministry of the Interior in 1997. In the perspective of the Augustus Method, the competences and the intervention areas are defined with a set of Support Functions, each of which has a specific Manager who is in charge of keeping their plan up to date, also through exercises and updates. In case of an emergency, the Support Functions are called to deploy their skills and are coordinated to respond effectively to the needs.

Support Functions are defined both at municipal and regional level. They include technical advisory and planning services, social and health care, veterinary services, media and information, volunteer organizations, traffic control and transport. In this way, the roles and responsibilities of the different operational centres are defined. The heads of the Support Functions actively participate in the operation centres, each for their own competence, ensuring cooperation and the availability of the resources at their disposal.

The structure of any emergency plan according to the Augustus Method has three basic parts: (i) General part, (ii) Outlines of Planning, (iii) Model of intervention.

The General Part includes all the information related to the knowledge of the territory, the existing monitoring networks, the potential risks in the area and the related possible incident scenarios (fires, floods, earthquakes, toxic smoke pollution). The Outlines of Planning part describes the objectives to be achieved in response to a specific emergency. Finally, the Model of Intervention part defines the responsibilities and the chain of command and control for all actions necessary to deal with civil protection emergencies. This also requires a constant exchange of information between the central coordination system and the peripheral system of the civil protection agencies about the updated availability of resources in terms of men and equipment to adequately deal with emergency situations.

A directive of the President of the Council of Ministers in 2005 further detailed the structure of the plan that in the new version includes the following sections: General Part, Incident Scenarios, Organisational Model of Intervention, Information to the Population, and Cartography.

In this article we report the results of the analysis of several emergency plans [10 , 26] that were selected because they represent different examples in terms of scope, size and responsible institution (author of the plan). These are reasonably recent plans, some of which have been updated over the last five years, and clearly we have considered only the most recent versions. In particular, we have focused on the active part of the contingency plan, i.e. mainly the Organisational Model of Intervention.

Interestingly, the considered plans show two different outlines for this section: some adopt a ‘phase-based’ structure, while others a ‘role-based’ one.

The document [26] includes the intervention plans for each alert level (Warning, Pre-alarm, Alarm) and the role of all the actors involved. Emergency management actions are described through a chronological description of the tasks and functions of all the entities involved in each alert level. This kind of structure sometime is not direct in identifying who has to fulfil each specific task, likely that descends from an attempt of avoiding repetitions in the text.

Instead, the plan reported in [12] describes which events trigger the transition from one alert level to another and then lists the description of individual roles and their responsibilities and actions to be implemented directly or to be activated in the form of a single list for each plan that discusses the tasks of the roles in all alert levels.

The second type of structure is found in longer documents, likely because in this way it is easier to locate the information concerning each stakeholder, thus facilitating the access to information in case of an urgency.

The distinction between these two types of structure is relevant to our study because we found that the two structures generate different types of ambiguity. While plans in the first category are sometimes vague about who is responsible for a specific task, plans in the second category eliminate this risk but are less clear in reporting the relationship between each alert level and the tasks to be performed withinthat.

3.2 Semantic approaches for emergency plans

A major problem in collecting, representing and integrating information from several heterogeneous sources can be simplified by an adequate conceptualization of the Emergency Management domain. An ontology, in fact, can provide a unified explanation of concepts and the relationships between them. Thus, it enables knowledge sharing among different users, and also it allows automatic data processing.

In the scientific literature, there are various studies and proposals of ontologies for the management of various types of emergencies (fires, explosions, terrorist attacks, natural disasters, humanitarian emergencies) which can have serious consequences on the well-being of the population. For example, [22] reports an extensive analysis of scientific articles in the literature. They discuss a classification of ontologies according to common concepts (people, organizations, resources, disasters, geography, processes, infrastructure, damage) and less common ones (topography, hydrology and meteorology).

In [19], the authors propose to integrate many ontologies and vocabularies found in the literature into a unified structure. The authors have followed the principles of ontological methodologies such as NeON [42] and Methontology [17] which encourage the reuse of existing ontologies. Furthermore, in order to have a quantitative assessment of the quality of their proposal, they designed an evaluation survey based on 17 questions concerning hierarchical, relational and lexical aspects.

Unfortunately, very few ontologies originally designed for crisis management are formally represented and accessible to the public. Moreover, there are no ontologies that cover all aspects of the emergency management domain and, above all, contingency plans are hardly ever written on the basis of an internationally accepted domain ontology.

The main issue that hinders the applicability of such ontologies in the context of the N.E.T.TUN.IT project is that they focus more on the description of the kind of event, resources and messages, whereas they lack in describing the structure of involved human organization, their goals, tasks and responsibilities.

This aspect is partially noticed in [3], where the authors present a metamodel for modelling a crisis situation from an analysis of the domain of interest with the aim of generating an interoperability layer (Mediation Information System, MIS) between the information systems of the organizations involved in responding to a crisis. In their metamodel, authors introduce Risk and Crisis elements. Risk models the conjunction of the possibility to have an event with negative consequences. The degree of risk is related to the probability of the event and its potential effects. A Crisis is the realization of a risk, and therefore it is related to the events and dangers of that risk. The presence of these elements is a direct consequence of the authors’ will to model a current emergency (a Crisis in their metamodel).

In [45], the authors propose an Emergency Response Organization ontology to overcome semantic ambiguities due to differences in emergency systems between different countries, regions and organizations. Authors focus on the emergency response organization with the aim of resolving semantic conflicts in the development of emergency response information systems or the elaboration of emergency plans.

This work goes in a direction that is promising for the N.E.T.TUN.IT project. They focus on (and detail very much) the concept of human organization and roles, but despite they introduce the concept of goal, that is quite marginal in their metamodel. According to our experience and project requirements, we aim at a better integration of goals/responsibilities with actors and tasks.

4 Linguistic and semantics

This section describes the semantic layer that is the ground for producing structured-text plans. This layer extends a preliminary work on lexical semantics (i.e. investigating word meaning) done in [6] and it refines the conceptual semantics by means of a metamodel, thus explaining properties of argument structure.

We started from a set of meaningful words identified through a linguistic analysis. A lexicon [2] is a representative set of words, where special words, called lemmas, became representative for a class of synonyms. Identifying lemmas and their meaning we re-create the missing link between the linguistic layer and the semantic one. The next subsections details this approach.

4.1 Linguistic analysis

One of the cornerstones of this work is defining the way in which a structured-text plan (a knowledge model) can be connected with its linguistic formulation, i.e. the free-text version of the plan. In practice, we need to analyse linguistic information thus to properly assign a specific meaning to words [2, 5].

To write the present work, we studied and analysed a number of Italian Emergency Plans [10 , 26] to compare the different approaches, styles and terminologies. The major issue we discovered is that each author uses to express the guidelines by using her own sensibility and linguistic knowledge; this often leads to the use of different words (synonyms) to address the same concept. Sometimes, the use of synonyms simply occurs for reasons of elegance and to avoid repetitions in close sentences.

In order to avoid an oversimplification of the problem, we think a good strategy would be to accept the richness of natural language, even if that means dealing with more complex inputs. In dealing with natural language documents, we delineate a clear separation between the linguistic and ontological levels, however, a close collaboration between them may help to better manage linguistic flaws (in particular, synonyms). The identification of lemmas in the lexicon is an instrument to connect linguistic and semantic layers [2]. In literature, a lexicon is defined as a vocabulary of meaningful units (words), in a specific language, along with the knowledge of how each word is used. A lemma is a specific word that raises as representative for a class of similar words. The concept of lemma gathers the linguistic properties of terms and their syntactic relations, differently from an ontology term that focuses on a conceptual element.

In order to identify these linguistically different but semantically similar words, we have defined a lexicon table (Table 1) as the result of a manual process of revision of emergency plans. Currently, it counts thirty-three terms referring to seven lemmas. In the first column, the reader can see the main word (i.e. the lemma) chosen as representative of all the terms (synonyms) listed in the second column. In the third column, we specify the precise meaning we associate to the whole class of words (lemmas and synonyms). Finally, in the fourth column, we show a sentence taken from an Italian plan, and translated into English, that contains one of the synonyms.

Table 1
Lexicon Table, containing the list of Lemmas and their Synonyms

Lemma Synonym Terms English (Italian) Meaning (for disambiguation) Example

Order Orders (Ordina) Give an authoritative instruction to do something The prefect disposes a continuous monitoring of air quality in relation to wind direction, intensity and altitude

Disposes (Dispone)

Provides (Provvede)

Sends (Invia)

Inform Informs (Informa) Give someone facts or information The Manager shall update the Prefect and other interested parties by means of messages using the template specified in the attachments

Updates (Aggiorna)

Communicates (Comunica)

Notifies (Notifica)

Signals (Segnala)

Reports (Relaziona)

Gather Data Monitors (Segue) Collect various types of information The mayor monitors the development of the situation and informs the population that the state of external emergency has been lifted.

Acquires (Acquisisce)

Observes (Osserva)

Carry out (Effettua misure)

Activate Activates (Attiva) Turn something to active or operative The Technical Rescue Director sets up an Advanced Command Post

Sets up (Istituisce)

Implements (Pone in essere)

Adopts (Adotta)

Actuates (Attua)

Applies (Applica)

Decide Decides (Decide) Come to a resolution in the mind as a result of consideration Mayor locates safe area for setting up a first accommodation facility

Locates (Individua)

Marks (Caratterizza)

Establishes (Stabilisce)

Defines (Definisce)

Authorizes (Autorizza)

Arrange Arranges (Predispone) Make instructions or preparations for an event or activity The mayor makes use of the municipal voluntary association

Coordinates (Coordina)

Organizes (Organizza)

Makes use (Si avvale)

Lemma	Synonym Terms English (Italian)	Meaning (for disambiguation)	Example
Order	Orders (Ordina)	Give an authoritative instruction to do something	The prefect disposes a continuous monitoring of air quality in relation to wind direction, intensity and altitude
	Disposes (Dispone)
	Provides (Provvede)
	Sends (Invia)
Inform	Informs (Informa)	Give someone facts or information	The Manager shall update the Prefect and other interested parties by means of messages using the template specified in the attachments
	Updates (Aggiorna)
	Communicates (Comunica)
	Notifies (Notifica)
	Signals (Segnala)
	Reports (Relaziona)
Gather Data	Monitors (Segue)	Collect various types of information	The mayor monitors the development of the situation and informs the population that the state of external emergency has been lifted.
	Acquires (Acquisisce)
	Observes (Osserva)
	Carry out (Effettua misure)
Activate	Activates (Attiva)	Turn something to active or operative	The Technical Rescue Director sets up an Advanced Command Post
	Sets up (Istituisce)
	Implements (Pone in essere)
	Adopts (Adotta)
	Actuates (Attua)
	Applies (Applica)
Decide	Decides (Decide)	Come to a resolution in the mind as a result of consideration	Mayor locates safe area for setting up a first accommodation facility
	Locates (Individua)
	Marks (Caratterizza)
	Establishes (Stabilisce)
	Defines (Definisce)
	Authorizes (Autorizza)
Arrange	Arranges (Predispone)	Make instructions or preparations for an event or activity	The mayor makes use of the municipal voluntary association
	Coordinates (Coordina)
	Organizes (Organizza)
	Makes use (Si avvale)

Discovering synonyms is significant because it limits the arbitrariness in interpreting the free-text plan, and it also allows moving from it to the structured-text version of the same plan. Moreover, it aids in creating linguistic categories that converge in the specification of lemmas [5] that are the representatives for all the contained words.

The Table 1 could grow, in the future, in order to better cover all the particular cases that yield in the whole corpus of national emergency plans. To this aim, it will be necessary to consider a larger number of emergency plans with the purpose of extending the current list of synonyms for each class of meaning, thus gradually reducing possible cases of ambiguity.

However, besides the problem of synonyms, that we partially solved here, we also encountered other linguistic issues that we noted for future investigations and are listed in Section 6. Given the objective to manually convert a document from a free-form text to a structured one, in most of the cases, humans are able to solve these flaws and take the correct decision by deducing the correct interpretation from the context. Indeed, these flaws represent a big limit when moving from human to machine interpretation of the text. After creating a semantic layer, the next natural step that comes to mind is moving towards an automatic support for extracting an ontology from the input text. Natural language processing is a fast-paced research field that may provide an important improvement to the automatic management of emergency management [8 , 44]. This is actually out of the scope, but we are moving forward with this objective in mind.

4.2 The proposed metamodel

It is now necessary to clarify some rules we adopted in defining the elements of the metamodel:

We are interested in lemmas because we want to connect the use of the metamodel to linguistic information embedded into the free-text plan [5].

We focus on a general structure of the emergency response plan: this means that domain actions related to the management of a specific accident (even if common to other cases) are out of the scope. The reason for this choice is to limit the number of elements of the metamodel and to remain general by leaving apart domain-dependent terms that change with the kind of emergency or that may depend on the adoption of new strategies and new technologies.

Typically, the elements of the ontology explicitly appears in the free-text version of the plan as lexicon. Our idea is to respect the problem knowledge and comprehension that the writer of the plan has. However, we extended the metamodel with some terms, e.g. decisions, and responsibilities, that do not explicitly appear as words in the plan’s text, but they are deducible from the document’s structure. An example of that will be provided in Section 5.

The rest of this section discusses the proposed metamodel, whereas the next section will put the metamodel in practice for building some parts of a structured-text plan.

In literature, many ontologies exist for each specific category of emergency (natural disasters [1], explosions [22] and terrorist attacks [23], just to mention some of these). As already discussed, our proposed semantics aims at highlighting collaboration patterns that appear within an emergency plan. For this reason, it does not focus on domain-specific actions used to solve some kind of accident rather than another one.

From the linguistic analysis, we discovered several elements that could be clustered in five categories: actors, events, actions, responsibilities and resources [6]. These elements constitute the basis for building a metamodel that enriches the previous work by detailing and relating these categories. This approach ensures a link to the linguistic level.

Actors are relevant because of the human-intensive nature of the response organization. Events yield to be modelled because they trigger changes in the current situation, delivering the different stages of the plan (for instance, moving from pre-alert to alert). Actions are of paramount importance in the metamodel because they represent the building block of any emergency plan. In the vision of adopting a goal-oriented self-adaptive systems, identifying responsibilities as goals is necessary for studying how the system will adapt to unexpected accident evolution. To this purpose, responsibilities specify why a specific actor is involved in the emergency management plan, what she should care of and pursue, and who could be considered accountable for a possible failure. A specific attention is given to delegation (term that originated from goal-oriented methodologies), that here refers the ‘principle of expected result’. Finally, resources are relevant, in the context of an emergency, because they provide means for addressing objectives prescribed in theplan.

In the following subsections, we will detail the fundamental aspects of these categories, and of the elements within them.

4.3 Actors

A fundamental aspect of the innovation proposed by the Augustus method consists in clearly assigning responsibilities. For this reason, it becomes very relevant to create a list of actors involved in the execution of the plan.

In the numerous plans we studied, we found many actors, often referred to with acronyms. We noted that the ambiguity allowed by the Italian language sometimes creates some indecision on identifying who is the actor responsible for performing a specific action. This indecision mainly happens in plans where the description is ordered using time or event-related criteria. Plans, where activities are clustered according to actors, do not present this ambiguity, of course.

We differentiate between individual and collective actors. With the term individual actor, we will address the common-sense meaning of a participant in an action or process. Collective actors represent a more refined concept where according to [37] collective actors perform a coordinated and collaborative decision-making process where one individual speaks for the group. Collective actors share the same interests, integration mechanisms, an internal and external representation of the collective actor and an innovation capacity.

Examples of individual actors include some already cited authorities: Mayor, Prefect, chairs and participants of committees (that are collective actors), for instance, the Responsible for the Town Operating Center or the Civil Protection Officer on duty.

Examples of collective actors include the operation room of the Metropolitan Police, the Rescue Coordination Center, the Regional Agency for Environment Protection, the Integrated Regional Operation Room, and so on.

As reported in the metamodel (see Figure 3), an actor may perform actions on its own (as the owner of the action), or it may assist another actor; this entails that the actor is the owner of responsibilities associated to the result of the action, i.e. she is accountable for the expected result; emergency plans lie in a knowledge-intensive field, for this reason the metamodel prescribes that the actor owns some knowledge useful for performing its duty. Another interesting aspect is that the action may be performed (by the owner or assistant) on behalf of someone else (after a delegation), finally, the actor may be either a producer or consumer of events.

Fig. 3

The proposed metamodel for emergency plans.

4.4 Responsibility

An emergency plan assigns specific responsibilities to the participant actors. For instance, according to the Augustus Method, each Support Function manager is in charge of a specific responsibility such as ensuring health-social assistance, managing mass media and information, coordinate voluntary organizations, controlling circulation and traffic, and so on. A Responsibility implies a commitment to address an objective under a personal responsibility and accountability also by law. Other examples of responsibilities are:

during the Alarm Phase, the Chief of the Provincial Fire Brigade is responsible for coordinating the technical and scientific staff;

the Provincial Health Agency General Manager is responsible for activating the necessary organisation for the specific type of accident;

the Chief of the City Brigade Fire is responsible for coordinating all operative structures forming the Rescue Coordination Centre.

We implicitly related the concept of responsibility to that of Goal because we intend to create a correspondence with strategic actor relationships that originate from social modelling [16] and Goal-Oriented requirements engineering [43] and to pave the way for a multiagent system automatic support [7]. In this perspective, responsibilities lead to either direct actions or delegations to other parties.The metamodel expresses this aspect by specializing the generic concept of delegation in two subclasses: Entrusting moves one responsibility from an authority to a subordinate actor, thus the latter becomes accountable for the expected result. Conversely, an Assignment implies that an activity is delegated, whereas the responsibility remains to the authority actor.

An interesting relationship is ‘means-end’ that connects a resource to a responsibility, thus modelling that the related resource is needed to fulfil the given responsibility.

4.5 Events

Emergencies exist because negative events compromise the environment, changing its state; this is a simple fact that justifies the importance of modelling events in an emergency. In our analysis, we always found that events are produced or consumed by actors (representing a sort of knowledge of the external event).

We identified two types of relevant events for the structured representation of the plan: Data Events and Messages. A Data Event represents data obtained from monitoring activities and the acquisition of information from any possible source. The Message event represents an intentional exchange of data (see the Inform action) that is related to the emergency, for instance, a phone call from the responsible manager of a plant affected by a significant fire blast.

Everything can be the content of a Message Event, however for simplicity we classify two subcategories of message: Informal and Formal. Informal messages refer to phone calls, media diffusion of news, and so on. The essential feature of an informal message is that it does not have any kind of template specified in the emergency plan. Formal messages are delivered using traceable communication means (emails, other types of computer-based messages, telegrams,...). An essential feature of formal messages is the adherence to a format specified in the plan. Frequently, formal messages are encoded using some emergency communication protocol, like the Common Alert Protocol (CAP) [30].

4.6 Actions

Actions are the essential brick of an emergency plan. The Action is an abstract category that is specialized by different subclasses delivering different organization patterns (that we identified in the linguistic analysis): order, activate, arrange, gather data, inform, decide.

All the actions are related to an owner actor who is up to execute them (sometimes on behalf of an authority), and optionally, some assistant actors helping in addressing the result. In the following, we discuss the subclasses of the Action abstract category.

Order . The first subclass we specialize from Action is Order, suitable for situations in which an authority actor gives an order to a recipient actor. The order typically generates a delegation (either the Entrusting of a responsibility or the Assignment of a duty).

Examples from plans we examined are:

The Prefect orders to the commissioner the actuation of the traffic deviation plan.

The Mayor orders the police to evacuate the zone.

Activate . Another subclass is Activate, it addresses situations where a resource is made available for being used within an emergency. Here, resource is a general term used to indicate three categories of elements that may be activated: a Plan (typically a sub-plan ready to be enacted), an Asset (e.g. an individual, an office that provides a function, a machinery, and so on) and a Knowledge (e.g. a document, a norm/law, some kind of information source).

Examples:

The Civil Protection Office Head activates the weather monitoring team.

The Emergency Manager activates the External Emergency Response Plan.

The Head of the Civil Protection Agency activates its Cartography Support Service.

Arrange . We also considered the Arrange subclass that implies some kind of coordination, organization and planning before a Plan may be enacted. This is a frequent occurrence in the emergency plans we studied. Examples:

The police arranges the monitoring of the emergency site to ensure the fast evacuation of the population.

The head of the Town Civil Protection Agency arranges watch duties in the Town Operation Room.

Gather data . The Gather Data subclass refers to the act of acquiring data, and it is very significant in dynamic emergency responses.

The Gather Data action is always performed over the current situation (data source) via direct/indirect observation, by remote sensors, experts operating in the field, and, sometimes, by citizens. It is also used for checking the operational status of resources.

Examples:

The Regional Civil Protection Agency gathers data about the accident.

The head of the Resource Department gathers data about the status of resources and personnel.

The head of the Civil Protection Agency gathers data about suitable places to shelter displaced persons.

Inform . Inform represents the act of sending messages to a recipient. The message may be a formal act, following a structured protocol, or an informal communication like a phone call. The action may be one-shot or repeated with a frequency (e.g. when updating someone about the current state of the monitored site).

The head of the Civil Protection informs the commander of the Municipal Police about the emergency status

The mayor informs the population about the incident in progress

Decide . Decide is a type of action requiring information, quick access to knowledge (such as maps, technical schematics of plants, norms) or expertise (i.e. support from technical staff). The output is a Decision, i.e. the (typically formal) act that has consequences over the plan. Consequences are generally provided in terms of other actions to be performed.

The Mayor decides to move from the pre-alert phase to the alert phase

The Civil Protection expert decides what are the needed maps for the management of the emergency

Decisions in the plans are often described in terms of the actor who has to take them and the possible alternatives (i.e. trigger alert, or deactivate the pre-alert phase). Sometimes some supporting actors are also listed. Decisions are the part of the plan that we have often found lacking relevant details; for instance, criteria for deciding may remain blurred and are rarely formalized. Often, plans do not explicitly use the ‘Decide’ lemma; they instead address the concept of a decision to be taken by someone by describing the incoming events and expected decisions in terms of orders issued or actions undertaken. Sometimes, events may be related to the emergency development, and decisions are about the actions that are required to face the new event.

5 Extracting emergency plans

This section presents the method we conceived for manually analysing a free-text emergency plan with the aim of generating its structured version. The section exploits several excerpts from the studied emergency plans, also presenting the corresponding instance of the metamodel.

In the following list, we report the steps used to transform emergency plans from free-form text into structured text. We suppose to fragment the plan in sentences or group of sentences that are logically related in terms of actions to perform, objective to achieve, timeframe specification, and so on. Each fragment will be processed with the following procedure:

Highlight candidate keywords (subjects, verbs, complements).

Identify actions (verbs of the metamodel or their synonyms from the Synonyms list) in the highlighted keywords.

Verify that actions used in the fragment match the semantics adopted in the proposed approach by considering that:

The semantics underpinned by the metamodel (see Fig. 3) is the same (particularly significant are the relationships with other metamodel elements).

Verbs in the analysed fragment of text are used in the same way the corresponding actions are used in proposed templates (see Appendix B).

Items listed in the property table of the verb (see Appendix A) match what can be extracted from the studied fragment (carefully verifying the presence of all the required properties).

Examples reported in the Synonyms list (Sect. 4) address the same meaning.

Rewrite the fragment of text by using the templates proposed in Appendix B (the result is the structured text version of the fragment).

Compile the property table of all the metamodel elements used in the structured text.

Verify that all words in the fragment have been properly considered and converted to structured text if they are worth of.2

The application domain of the proposed metamodel is quite specific (civil protection emergency plans), so it is possible to empirically verify the coverage of the metamodel with respect to its scope. Moreover, despite emergency plans follow norms and laws that can vary significantly in different nations, the elements presented in our metamodel seem general enough to cover most of the cases. Within the N.E.T.TUN.IT project, we will have the opportunity to compare this metamodel with plans used in the countries of other partners, so to obtain a further validation. Indeed, this is one of the reasons we preferred to identify only a few (quite general) lemmas and to complete the coverage with a list of synonymous. It is sufficient to map the proposed lemmas to those used in another language for having a reasonable starting point for the analysis of a plan in another language. A similar issue regards the actions that are specifically adopted for countering each different accident. They are accident-specific and of limited interest for the scope of this paper, but they can easily be deducted from the reading of each specific plan and identified throughout that. The reuse of these terms in other plans is, anyway, limited when we change the accident at hand.

We will now provide a few examples of sentences extracted from real emergency plans, more sentences are reported in Appendix C. The sentences are originally in Italian, translated in English trying to maintain the original meaning, but this may introduce some flaw. Examples are introduced adopting a precise structure:

The rationale behind the selection of the sentence as an example, we also cite the document where the sentence has been found.

The sentence in natural text, as found in the emergency plan.

An analysis of the text and some considerations arising from that.

The instance of the proposed metamodel springing from the sentence.

The properties tables for the metamodel elements identified in the sentence.

the structured text version of the sentence obtained by using the templates reported in Appendix.

Example 5.1. The objective of this example is to smoothly introduce the reader to the proposed method. The following fragment of plan, actually a simple sentence extracted from [11, p. 42]palermo, is a case of Actor-Action-Object phrase with a straightforward interpretation.

Sentence 1 — “The municipal police arrange the assigned roadblocks”

This sentence is characterized by a subject (the municipal police) a verb (arrange) and a noun (the roadblock). The Arrange action is a lemma of the metamodel that refers to something that must be planned/organized/adapted at run-time. This perfectly matches with the very nature of putting roadblocks in the area to be secured. Fig. 4 shows an instance of the portion of the proposed metamodel, representing these three elements and their relationships.

Fig. 4

Instance of the metamodel for Sentence 5.

By using the property table of the element Arrange (see the Appendix), we analyse the input text for further details. Table 2 shows a minimalist table where only non-trivial rows are filled with values (for convenience, we omitted optional fields).

Table 2

Properties of Arrange activity for Sentence 5

Arrange
owner	Municipal police
object	Assigned roadblocks

Finally, from the Appendix B, we also get that the sentence template for the Arrange element is something like: [owner] Arranges [object].

Therefore, we positively verified that the Arrange action is used in the meaning addressed by our metamodel and now, by using the proposed sentence template, we obtain the following output structured sentence:

Structured Sentence 5 — “[Municipal police] arrange [Assigned roadblocks]”

Example 5.2. The next sentence is a more complex case; it is extracted from [13, p. 93]siracusa, and it includes two actors, two actions, and an event that will be decomposed into two correlated parts.

Sentence 2 — “The head of the civil protection agency alerts the director of the volunteer service for the activation of volunteer organizations from neighboring provinces.”

Sentence 5 involves two actors: the head of the civil protection agency and the director of the volunteer service. The former pursues the primary activity of ordering something, the latter pursues the order by activating a resource. The order generates an assignment. When an order is dispatched, a duty is assigned, in this case the activation of the volunteer organizations. We model this using the ‘Order-Assignment-Activate’ block. The director of the volunteer service owns the assigned activity ‘Activate’. Fig. 5 shows the instance of the metamodel for Sentence 5.

Fig. 5

Instance of the metamodel for Sentence 5.

Table 3 lists the properties related to activity ‘Order’ in Sentence 5 (for convenience, only non-optional fields):

Table 3

Properties of Order activity for Sentence 5

	Order
owner	Head of civil protection agency
generate	Activate
target	Director of volunteer service

Table 4 lists the properties related to activity Activate in Sentence 5 (for convenience, only non-optional fields):

Table 4

Properties of Activate activity for Sentence 5

	Activate
owner	Director of volunteer service
object	Volunteer organizations from neighboring provinces
on-behalf-of	Head of civil protection agency

According to the templates the activities characterizing Sentence 5 (see Appendix), the instance of the metamodel is the following:

Structured Sentence 5 — “[Head of civil protection agency] orders [Director of volunteer service] to [activate] with assignment [Volunteer organizations from neighboring provinces]. Consequently, [Director of volunteer service] activates [Volunteer organizations from neighboring provinces].”

6 Experimental results and open issues

This section presents an experiment to validate the proposed approach, and discusses limits and open problems.

6.1 Experimental phase

We set up an experiment to evaluate the appropriateness of the proposed approach. In the following, we want to analyse the linguistic and semantic layers for the purpose of translating some target emergency plans with respect to validity from the point of view of a potential user 3 in the context of researchers.

The experiment has been designed to evaluate three aspects:

RQ 1: the completeness of the lexicon table;

RQ 2: the coverage of the metamodel;

RQ 3: the usefulness of the tables of property.

To respond to RQ 1 we asked to annotate all the words that do not match with any of the identified lemmas and synonymous of the lexicon table. We also asked to double-check (and count) possible wrong associations (occurrences of polysemy).

To respond to RQ 2 we measure the cases in which the metamodel is sufficient to model the target sentence.

To evaluate RQ 3, we asked subjects to respond to which extent filling the table of properties helped in the disambiguation of an element of the metamodel.

In order to conduct the experiment, we selected a team of 4 researchers. Two of them were not involved in the research phase (in which the linguistic and semantic layers were build up), therefore their knowledge about the proposed approach (and of the emergency domain) was limited.

Therefore, we selected two documents containing two emergency plans of different type. The first document [13] is 120 pages with about 44,000 words. The second one [12] is 45 pages with about 35,000 words.

Given the input document, we designed the experiment operationalization by two tasks.

Task 1 - From these documents, the first task was to identify the relevant sentences that describe some kind of reaction to an emergency event.

Task 2 - Translating the sentence by using the proposed approach.

We can classify the experiment as 1 factor with 2 treatments [46], given that the same task will be operated by a subject with background knowledge and by another subject without background knowledge.

The experiment run occurs in two different moments:

Stage 1 - all the subjects were involved in teamwork for processing documents to identify the relevant sentences (Task 1). After completing this task over the two documents, the list of sentences is randomized and distributed to the 4 subjects.

Stage 2 - each subject, working individually, processes its own queue of sentences in order to instantiate the metamodel (Task 2). This stage is double-checked, in the sense that every completed sentence is published in a shared document, where all the other subject can access. To complete the stage, every produced artefact must be revised and marked as reviewed.

We did not assigned limited time for the exercise, and we let subjects to work in different moments. The only synchronization point was the end of stage 1 and the beginning of stage 2.

From the experiment running, we obtained some preliminary findings about the quality of the proposed approach. The most important result concerns RQ 2, i.e. the completeness of the metamodel. Indeed, from a semantic point of view, the coverage was 100% : all the cases of the input documents were successfully instantiated by the metamodel. Therefore, we can reject null hypothesis and accept RQ 2.

A different result occurred for the linguistic layer. Subjects annotated different cases of words that did not match with any lemma and synonyms; moreover, they discovered many cases of polysemy. We can not conclude about the completeness of the lexicon table and must reject RQ 1. The linguistic problems were extensively analysed, and we discuss them in the following subsection.

Finally, the four subjects fully agree about the usefulness of the table of properties as a complementary tool for deducting the semantic structure of a sentence. Consequently, RQ 3 is accepted.

For what concerns the validity of the experiment, we can surely refer about the low statistical power due to a limited number of participating subjects, that is partially compensated by a large size of input documents. Moreover, the usefulness of the tables of property is demanded to a subjective perception of the involved personnel that may suffer of the fishing and error rate. A factor that may affect the internal validity of the experiment was the not synchronicity of scheduled tasks, due to personal agendas. This modality could have generated some interactions among personnel, therefore treatments could have influenced each other.

6.2 Open linguistic problems

The approach we propose in the paper has several advantages: it allows disambiguation of the text, it enforces the clear identification of responsibilities, and the delegation of tasks/responsibilities. The structured form of the text may be used for the generation of a workflow corresponding to the plan. Connecting the semantic layer to a linguistic analysis produces a well-grounded set of lemmas that are represented into a metamodel. However, practical usage of the semantics is actually hindered by a number of linguistic issues that we discovered, but not all of them are yet solved. In the following, we provide a list of linguistic challenges that we raise, in order to move from a manual to an automatic textconversion.

Another relevant consideration is that, despite the fact a semantic layer reduces the effort required to manually convert the free-text form of the plan to the structured one, we are aware that a manual conversion is too demanding. For this reason we are also considering the adoption of an automatic text conversion approach on the basis of existing contributions in the literature like [8 , 44].

In the following, we provide a classification of some linguistic issues that we encountered during the preparation of this work.

Polysemic words. Words in this category have several possible meanings [36]. Let us consider the following example: “The head of the Civil Protection Agency identifies suitable places to shelter displaced persons”; it shows the use of a synonym (’identifies’ in place of ’gathers data’) in the proposed sentence. This is an intriguing case of ambiguity that may be solved through the context. Indeed, the term ‘identify’ could also be intended as a decision to be taken about what are the best places to shelter displaced persons. However, we deduced that the sentence refers to a ’gathers data’ act by looking at the rest of the sentence, where the delivered meaning is to collect geographical data for future use.

Another frequent case of polysemy concerns the word ‘alert’ that can be used with the intended meaning of Informing someone about an event, or it can be used to express an Order as reported in Sentence 5 (see previous section).

The problem of polysemy is frequent when interpreting natural language text, and it is amplified in an emergency plan due to the super specialized vocabulary that is used with the purpose of writing plans. This is a big issue to the objective of automatic translating the text. State of the art in automatic sense disambiguation is to date considering this a task of immense difficulty [38]. An interesting claim comes from [15] where the author observes that polysemy is easier to handle at a conceptual level, and proposes a frame-based methodology that exploits domain ontologies for reducing the negative effects of polysemic words. {% } In this approach, polysemous unit are distinguished thanks to the conceptual relations established with other concepts belonging to the same communicative situation.

Composed Meaning. We also discovered some particular cases in which the same term refers to a composed meaning, i.e. it may be expressed by composing other lemmas. An example is the term “evaluate”; let us consider the sentence:

“Fire brigade evaluates possible need for resources”

this one corresponds into the following actions:

Fire brigade gathers data about existing resources;

Fire brigade decides on the need for additional resources.

In the example above the word in the free text “evaluates” requires two successive actions which in our structure are represented by the lemmas “gather data” and “decide”.

To the best of our knowledge, literature does not report contributions that identified such a category of linguistic issues. Our agenda already contains an empirical study to be performed over many emergency plans with the intent of discovering the highest number of words with a composed meaning.

Normalization via pre-treatment. There exist words in some sentences of emergency plans that at a first analysis seem to be uncovered by the proposed metamodel. Let us consider the following sentence: “The Municipal Police implements the directives of the Commissioner in any Public Order activities”. The metamodel does not support the action Implement. The term is reported among the synonyms of Activate in the Lexicon table (Table 1), however, ‘Activate the directives’ does not make sense, and it clearly does not match with the intended meaning of the sentence.

This actually is not a limit of the approach, but rather it is a limit of the linguistic analysis. Indeed, the same sentence could be rewritten as “The Commissioner orders the Municipal Police to implement …”. In this case, the sentence could easily be treated by our approach, because the Order lemma is well suited into the metamodel.

We classified this issue as a problem of ‘pre-treatment’ of the input text. We are convinced that several rules could be defined to solve problems in this category. For instance, switching subject and direct object of the previous sentence allows discovering that the Implement action is related to an Order.

Another relevant example concerns the Assist verb that is not covered by the lexicon. However, the metamodel includes Assistant as one of the properties of Action. This means that a sentence like “The Forestry Corps cooperates with the Fire Brigade in extinguishing a fire” could be normalized as “Fire Brigade have the responsibility to extinguish a fire with the support of The Forestry Corps”.

We are currently working on identifying a set of cases and the corresponding rules to be applied for obtaining a normalized version of a sentence. This is an ongoing work that has already provided preliminary results, and it will be completed in futureworks.

Relationships between sentences. As a matter of fact, a plan is not composed of a mere list of atomic sentences, rather, it is ideally a workflow including alternative branches, and, within them, sets of instructions that could be executed in parallel, in a specific order (mandatory or not), and according to predefined temporal constraints (also including temporal intervals/deadlines).

Usually, such specifications are delivered, in the free-text form of the plan, by using elements of the text (such as adverbs), or text structures (like numbered/dotted lists, and so on).

Disambiguation of such constraints is often a hard task. Just to provide an example, let us consider a bullet list of actions to be performed to mitigate the consequences of some accident (a very common situation in many plans). Is the list prescribing a mandatory order, or may the actions be executed even in parallel? We often found this is left to a common-sense interpretation of the work to be done, and therefore it strongly depends on the knowledge and skills of the reader. It is always worth remembering that in stress conditions such abilities may be altered and this may lead to a misinterpretation of the plan.

Again, we collected a set of examples, and we are deducing some cases that could help in identifying some patterns. We also envisage the need to establish a set of connectors that will be used in the structured version of the text to clearly specify relationships between sentences like: sequence, parallel, time dependencies, deadlines, and so on. Nonetheless, similarly to previous listed linguistic issues, this point remains an open challenge that is worth further studies.

7 Project roadmap

Worldwide, emergency plans are written in free-text form; this work arises from the need to generate a common and stable understanding of the organizational patterns that are described along several pages of text with all the flaws of natural language.

However, this is the first step of a pipeline for transforming a free-form text into an executable dynamic workflow. Indeed, in the context of the N.E.T.TUN.IT project, we conceived a three-steps process to address this long term objective (see also Fig. 6). The whole pipeline, together with the envisaged challenges, are described below.

Fig. 6

The proposed process for transitioning from free-text emergency plans to executable plans.

From Free-Text Plan to Structured-Text Plan, that is the focus of this paper. The use of a linguistic and semantic layers is the first step for making explicit the intended meaning of sentences.

Challenge 1 . Manual translation is clearly a great burden for the persons who will manually translate these documents. To facilitate their task, we are investigating the use of semi-automatic text conversion techniques from literature [8 , 44], and the evaluation of text complexity for identifying specific parts of a text that can be simplified to support the conversion process from free-form text to its structured version [40]. Surely, instruments like WordNet [27] will play a role in the semantic tagging of input text, whereas other instruments like Bablenet [28] will be useful in the phase of extending the approach to other languages (for example in French).

Representing the Structured-Text Plan with a Modelling Notation.

The first outcome from having a structured text, is the possibility to produce a graphical notation.

This point has been preliminary explored in [29], in the context of Norwegian emergency management. The authors show how authorities and rescuers better understand plans expressed in visual and textual form, and therefore, they can be more proficient in facing unanticipated events. This study also focuses on highlighting roles in the organizations and how they have to interact. The same research recognizes some problems using the BPMN standard: some difficulty to model task duration and in reusing a process diagram from one environment to another.

Our agenda includes devising a specific modelling notation to represent an emergency plan. An essentialrequirement for this notation is that it will easily support adopting an adaptive middleware for the execution and coordination of the plan’s activities. The definition of this notation is still a work-in-progress activity, but we have identified the main contributions it will receive from a few well-known standards. The BPMN notation [31] is a part of that, but not so central as it could be expected. The Italian directives (the Augustus method) prescribe that a plan provides general indications for the management of the emergency, whereas details are to be defined at emergency time. For this reason, we are considering to adopt the Case Management Modelling Notation (CMMN) [32] that allows to represent scenario-based situations, and the Decision Modelling Notation (DMN) [33] that allows to formalize critical aspects of decisions to be taken during the development of an accident, also including decision criteria (like data values reported by personnel on scene) and the reference documents to be consulted. Finally, some parts of the free-text plans naturally convey the opportunity to introduce a model of the goals related to the responsibilities of involved stakeholders (like the authorities and the support functions described in a plan).

Classically speaking, the literature broadly promotes supplementing natural language with standard notations and languages for business processes, such as the Business Process Modelling Notation (BPMN) [31]. However, designing high-quality emergency response processes is a great challenge that involves a relevant domain knowledge and the adoption of ad-hoc process modelling techniques. Indeed, BPMN has some limitations when applied to this specific domain, some coming from the intrinsic generality of emergency plans (the details of the accident are not known when writing the plan), others from domain-specific issues (like the relevance of location-related information and the employment of multiple resources [4]).

Challenge 2 . A possible to solution seems to be in complementing BPMN with other notations, namely the Case Management Model and Notation (CMMN) [24], and the Decision Model and Notation (DMN) [34]. In fact, while traditional business processes can be described by a priori defined sequences of activities using the BPMN notation [31], the CMMN notation (another OMG standard) offers a more natural support for dynamic workflows, while DMN (another OMG standard) allows describing and modelling repeatable decisions withinorganizations.

This position is also supported by [41], where authors highlight that an emergency response is a knowledge-intensive process, modelling and automating such a process is therefore a challenging task. Authors suggest to use CMMN and build a template model for a generic emergency response process. The adoption of the three modelling notations (BPMN, CMMN, DMN) at the same time is also suggested in [21], where the authors investigate how to use a combination of these three modelling languages in the context of crisis management.

Adaptive Execution and Management of Emergency Plan. We intend to automatically produce executable workflows to coordinate the activities within an emergency plan.

Challenge 3 . This is the step of our process towards the introduction of an automatic support for the management and execution of emergency plans is to work on a formal graphic notation that could well represent emergency processes. The idea is that the structured-text document will be the input to model the plan by using a notation that could be injected in some workflow management system, and automatically executed.

The definition of the best notation for such a task is a relevant challenge that is still open, and it will be part of our future research work.

Challenge 4 . The context of emergency management is extremely variable and very difficult to forecast. Unexpected events may spoil the validity of pre-defined plan: the actual plan ought to emerge after the consideration of the specific accident, environmental conditions (for instance road-weather conditions and so on), assets (fuel, food, transportation and so on), operational capability (assets, numerical consistency of assets, intervention time and need for logistic support).

An accident management system has to include relevant features of adaptive workflows (like MUSA [6, 39]) considering the roles of humans, environment data (spatial-referred representation of the environment and involved assets) and finally, laws and regulation.

Just to provide an example, the communication capability is a key element of any emergency. The support to emergency communications is already existing, and great care is devoted to the adoption of standardized content protocols for messages, as the Common Alerting Protocol (CAP) [30] that is an XML-based data format for exchanging public warnings and emergencies. Although that is of relevant value, more is still to be done on the telecommunication infrastructure resilience and the automatic support for alternative delivery channels.

8 Conclusions

This paper focused on inconsistencies, redundancies, and ambiguities that hinder understanding and formalizing emergency plans. The need to convert informal documents to more rigorous conceptual models requires a semantic layer for identifying essential elements of the input text and resolving linguistic issues that may be present. To achieve this result, we extracted essential keywords through an empirical study of several Italian documents reporting different kinds of emergency plans. Our analysis allowed us to discover recurrent structures in these documents. Sometimes, these linguistic structures are evident, other times they are hidden, and some interpretation of the text’s meaning is needed. We support identifying them by using a metamodel and by a list of properties for each lemma. The translation into the structured form of the text is supported by specific templates.

Footnotes

Acknowledgments

The research was partially founded by the DSB.AD008.645 "Net de l’Environnement Transfrontalière TUNisie-ITalie (N.E.T.TUN.IT)" research project, within the Cross-border Cooperation Programme of the EU Community, Italy-Tunisia 2014-2020.

Supplementary material

The appendix is available in the electronic version of this article: .

Net de l’Environnement Transfrontalièr TUNisie-ITalie (N.E.T.TUN.IT), Cross-border Cooperation Programme of the EU Community, Italy-Tunisia 2014-2020 (extended to 2023), project started on 03.09.2020 and concluding on 31.12.2022,

It is worth noting that the proposed approach still does not manage some portions of the plan text, for instance adverbs expressing causal/temporal relationships (we plan to deal with them in a future work), and accident-specific actions and terms (too many and too specific for being part of the metamodel).

In our vision, the potential user is a person with domain knowledge and IT skills, she is able to use a conceptual model, and she is able to understand an emergency plan.

References

Babitski

, Probst

, Hoffmann

and Oberle

, Ontology Design for Information Integration in Disaster Management, Informatik 2009–Im Focus das Leben, 2009.

Badra

, Despres

and Djedidi

, Ontology and Lexicon: the Missing Link. In Proceedings of the Workshop “Ontology and Lexicon: new Insights", Workshop at TIA 2011, 2011.

Benaben

, Hanachi

, Lauras

, Couget

and Chapurlat

, A Metamodel and its Ontology to Guide Crisis Characterization and its Collaborative Management. In Proceedings of 5th International ISCRAM Conference on Ontologies for Crisis Characterization May 2008 (2008), pp. 189–196.

Betke

and Seifert

, BPMN for Disaster Response Processes, INFORMATIK 2017, 2017.

Buitelaar

, Cimiano

, Haase

and Sintek

, Towards Linguistically Grounded Ontologies. In European Semantic Web Conference, (2009), pp. 111–125. Springer.

Cossentino

, Lopes

, Sabatucci

and Tripiciano

, Towards a Semantic Layer for Italian Emergency Plans. In Proceedings of the 22nd Workshop “From Objects to Agents", Bologna, Italy, September 1-3, 2021, volume 2963 of CEUR Workshop Proceedings, pp. 144–161. CEUR-WS.org, 2021.

Croatti

, Montagna

and Ricci

, A Personal Medical Digital Assistant Agent for Supporting Human Operators in Emergency Scenarios. In Agents and multi-agent systems for health care, pp. 59–75. Springer, 2017.

Cuzzocrea

and Pilato

, An Innovative User-Attentive Framework for Supporting Real-Time Detection and Mining of Streaming Microblog Posts, Soft Computing24(13) (2020), 9663–9682.

Dahab

M.Y.

, Hassan

H.A.

and Rafea

, TextOntoEx: Automatic Ontology Construction from Natural English Text, Expert Systems with Applications34(2) (2008), 1474–1480.

10.

di Palermo

, Piano Emergenza Esterna Esso Italiana S.r.l Deposito oli minerali Via Messina Marine 813 Palermo. Technical report, Prefettura di Palermo, 2016. Available at http://www.prefettura.it/FILES/AllegatiPag/1203/Piano_Emergenza_Estermo_Esso_AGGIORNAMENTO_ANNO_2016_-pubblicato-_.pdf.

11.

di Palermo

, Piano Comunale Di Protezione Civile - Relazione Generale. Technical report, Prefettura di Palermo, 2017. Available at https://www.comune.palermo.it/js/server/uploads/spc/documenti/_08092017103701.pdf.

12.

di Ravenna

, Piano di Emergenza Esterna per gli Stabilimenti di Soglia Superiore nella Provincia di Ravenna. Technical report, Prefettura di Ravenna, 2017. Available at http://racine.racine.ra.it/prefettura/newsito/pee/PEE_Cabot_CFS_ENI_Orion_Polynt_Rivoira_Vinavil_8set17web.pdf.

13.

di Siracusa

, Pianificazione di emergenza esterna per gli stabilimenti a rischio di incidente rilevante. Technical report, Prefettura di Siracusa, 2019. Available at http://www.prefettura.it/FILES/AllegatiPag/1146/PEEA_03.1_08102019.pdf.

14.

di Siracusa

, Pianificazione di Emergenza Esterna per gli Stabilimenti a Rischio di Incidente Rilevante. Piano Operativo Cancelli Tabella Generale Allegato 3.5.A. Technical report, Prefettura di Siracusa, 2019. Available at http://www.prefettura.it/FILES/AllegatiPag/1146/3.5.A._-_POC_Tabella_generale.pdf.

15.

Duran-Munoz.

, Managing Polysemy in the Adventure Tourism Discourse with Frame Semantics. In 9th International Conference on Terminology and Artificial Intelligence (2011), pp. 40.

16.

Eric

S.K.

, Giorgini

, Maiden

and Mylopoulos

, Social Modeling for Requirements Engineering, MIT press, 2011.

17.

Fernandez-Lopez

, Gomez-Perez

and Juristo

, METHONTOLOGY: from ontological art towards ontological engineering, Engineering Workshop on Ontological Engineering (AAAI97), 3 1997.

18.

Galanti

, Il metodo augustus, DPC Informa “Periodico informativo del Dipartimento della Protezione Civile", (4), 1997.

19.

Gaur

, Shekarpour

, Gyrard

and Sheth

, empathi: an Ontology for Emergency Managing and Planning about Hazard Crisis. In Proceedings of the IEEE 13th International Conference on Semantic Computing, 2019.

20.

Guo

, Zeng

, Duan

, Yuan

, Ni

and Liu

, Automatic Extraction of Emergency Response Process Models from Chinese Plans, IEEE Access6 (2018), 74104–74119.

21.

Herrera

M.P.R.

and Diaz

J.S.

, Improving Emergency Response through Business Process, Case Management, and Decision Models. In ISCRAM, 2019.

22.

, Shaw

and Brewster

, Ontologies for Crisis Management: a Review of State of the Art in Ontology Design and Usability. In Proceedings of the Information Systems for Crisis Response and Management conference, ISCRAM2013, 2013.

23.

Mannes

and Golbeck

, Building a Terrorism Ontology. In ISWC Workshop on Ontology Patterns for the Semantic Web, volume 36, 2005.

24.

Marin

M.A.

, Introduction to the Case Management Model and Notation (CMMN). arXiv preprint, 2016. arXiv:1608.05011.

25.

Marquez

, Carreras

, Litkowski

K.C.

and Stevenson

, Semantic Role Labeling: an Introduction to the Special Issue, 2008.

26.

Mereu

, Alferi

, Blandino

, Piano di Protezione Civile per la Salvaguardia della Popolazione dal Rischio Idrogeologico Derivante dalla Ostruzione del Canale Sotterraneo Papireto edizione 2015-16. Technical report, Comune di Palermo, 2015. Available at https://www.comune.palermo.it/js/server/uploads/spc/documenti/_08092017104149.pdf.

27.

Miller

G.A.

, WordNet: a Lexical Database for English, Communications of the ACM38(11) (1995), 39–41.

28.

Navigli

and Ponzetto

S.P.

, BabelNet: Building a Very Large Multilingual Semantic Network. In Proceedings of the 48th annual meeting of the association for computational linguistics, (2010), pp. 216–225.

29.

Nunavath

and Prinz

, Norwegian Emergency Management Process by Using Business Process Modeling Notation. In Proceedings of the 8th IADIS International Conference on Information Systems 2015, 2015.

30.

OASIS Standard. Common Alerting Protocol Version 1.2. http://docs.oasis-open.org/emergency/cap/v1.2/CAP-v1.2-os.html, 2010, Accessed: 01 July 2010.

31.

OMG. Business Process Model and Notation (BPMN), Version 2.0, January 2011.

32.

OMG. Case Management Model and Notation, Version 1.1, January 2016.

33.

OMG. Decision Model and Notation, Version 1.3, 2021.

34.

Object Management Group (OMG). Decision Model and Notation (DMN). Technical report, 2021.

35.

Palmer

, Gildea

and Xue

, Semantic Role Labeling, Synthesis Lectures on Human Language Technologies3(1) (2010), 1–103.

36.

Pesina

and Latushkina

, Polysemy and Cognition, Procedia - Social and Behavioral Sciences192 (2015), 486–490. The Proceedings of 2nd Global Conference on Conference on Linguistics and Foreign Language Teaching.

37.

Pichierri

, Lo sviluppo locale in Europa. Stato dell’arte e prospettive, Rubbettino, 2005.

38.

Ravin

and Leacock

, Polysemy: Theoretical and computational approaches, OUP Oxford, 2000.

39.

Sabatucci

, Lopes

and Cossentino

, MUSA 2.0: A Distributed and Scalable Middleware for User-Driven Service Adaptation. In International Conference on Intelligent Interactive Multimedia Systems and Services (2017), pp. 492–501. Springer, 2017.

40.

Schicchi

, Pilato

and Lo Bosco

, Deep Neural Attention-Based Model for the Evaluation of Italian SentencesComplexity. In 2020 IEEE 14th International Conference on Semantic Computing (ICSC) (2020), pp. 253–256.

41.

Shahrah

A.Y.

and Al-Mashari

M.A.

, Modelling Emergency Response Process Using Case Management Model and Notation, IET Software11(6) (2017), 301–308.

42.

Suarez-Figueroa

M.C.

, Gomez-Perez

and Fernandez-Lopez

, The NeOn Methodology for Ontology Engineering, (2012), pp. 9–34. Springer.

43.

Van Lamsweerde

, Goal-Oriented Requirements Engineering: a Guided Tour. In Proceedings Fifth IEEE International Symposium on Requirements Engineering, (2001), pp. 249–262. IEEE.

44.

Velardi

, Fabriani

and Missikoff

, Using Text Processing Techniques to Automatically Enrich a Domain Ontology. In Proceedings of the international conference on Formal Ontology in Information Systems-Volume 2001, (2001), pp. 270–284.

45.

Wang

, Zhang

, Dong

, Gao

, Du

and Lai

, Emergency Response Organization Ontology Model and its Application. In Second International Symposium on Information Science and Engineering (ISISE 2009), Los Alamitos, CA, USA, 2009. IEEE.

46.

Wohlin

, Runeson

, Host

, Ohlsson

M.C.

, Regnell

and Wesslen

, Experimentation in software engineering, Springer Science & Business Media, 2012.