This paper proposes a methodology for managing linguistic uncertainty in shared text. The methodology applies standard risk management practice and is illustrated in a scenario with stakeholders who make use of a shared reference text to make a decision that meets specified objectives. Methodology such as this is important in risk management practice because of stakeholders' inherent intolerance of uncertainty. An ontology that could be useful in artificial intelligence applications is also proposed.
Uncertainty is recognized as a source of risk, and one type is linguistic uncertainty. The objective of
this paper is to propose a methodology that applies standardized risk management practices to address
linguistic uncertainty. Methodology such as this is important in risk management practice because
uncertainty will otherwise likely go improperly addressed (and largely unaddressed). It has long been
known that people’s intolerance to uncertainty results in avoidance [
This paper begins by proposing a typology of linguistic uncertainty that is subsequently used in a
proposed methodology for managing linguistic uncertainty. Linguistic uncertainty is acknowledged
in established standards for risk management (e.g., ISO 31010 [
The use of this methodology is illustrated in a simple fictional scenario where service providers rely on their interpretation of shared text (i.e., their shared meaning) to choose among service oferings for each of their clients, and they must explain their choice. This illustration demonstrates how managing linguistic uncertainty could improve the consistency and explainability of choice of service oferings in such a scenario.
Further, a conceptual model and components of an ontology of Managing Linguistic Uncertainty
in shared Text (or, MLUT ontology) is proposed, and a distinction between the concepts of risk and
uncertainty is made evident. Components of this ontology are operationalized with reference to two
frameworks: (1) the Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE) detailed
in ISO/IEC 21838-3 standard [
The proposed topology of uncertainty is described next.
Hillen, Han and their colleagues analyzed a broad range of eighteen adopted measures of tolerance to
uncertainty in order to identify and categorize sources of uncertainty [
Some of these fourteen sources are identified by other impactful uncertainty researchers, including
those who have specifically studied linguistic uncertainty. For example, Divergence is described as
“conflict” by Benjamin and Budescu [
Hillen, Han and their colleagues describe their categorization as an integrative and flexible model that could be utilized in multiple domains and disciplines. While some of the sources of uncertainty seem similar, their distinction will become clear in the next section, where they are contextualized.
The fourteen (14) distinct sources of uncertainty identified in the last section are contextualized for linguistic uncertainty in shared meaning and described in the second column of Table 1.
Unpredictability, Impermanence, and Tentativeness all have a temporal dimension. However, while
Impermanence acknowledges the mutable nature of conceptual definitions, and the fact that the shared
text today might be interpreted somehow diferently in future, Unpredictability makes it challenging or
not possible to predict how the shared meaning (i.e., interpretation) of the text will change, or vary,
with time across Stakeholders. Regan and colleagues [
Tentativeness arises because of indecision in coming up with shared meaning. One feasible treatment to Tentativeness might be to revisit the shared meaning of the text to identify those aspects that remain undecided and to decide on them, and to document the remaining undecided aspects, reasons for them, and their potential impact to outcomes, which in this case are consistency and explainability of choice of service oferings.
Some sources of uncertainty may contribute to another source of uncertainty. For instance, equivocality (Polysmousness) in the shared text may create vagueness (Indefinitiveness). In such a case, addressing equivocality could improve vagueness. In contrast, while an obscurity (Non-Transparency) may also create vagueness, other obscurities may be present that do not. For this reason, addressing uncertainties should be done both separately and in combination.
The next section describes a methodology for managing linguistic uncertainty and illustrates this in a scenario with stakeholders (service providers) who make use of a shared text (a definition of homelessness) to make a decision (a choice of client service oferings) that must meet their objectives (consistency and explainability of their choice of service oferings). 1Hillen et al. (2017) uses the term uncertainty to refer to both uncertainty and ambiguity, and describes how this parsimony does not compromise the analysis. 2Note the distinction between Unpredictability of interpretation of the shared text homelessness and Unpredictability of the nature of homelessness of a client over time, which could also impact a service provider’s choice of the client’s service oferings. Additional uncertainty could arise from applying the interpretation of homelessness to predict the nature of homelessness of a client over time. This additional uncertainty would warrant additional management of uncertainty during the decision-making process.
This section will illustrate a methodology for managing linguistic uncertainty with a fictional scenario. In this scenario, service providers for clients experiencing homelessness make choices of service oferings for each of their clients based on a client’s needs, and providers must explain each choice.
In making their choice of service ofering, service providers rely on a shared reference document that includes a definition of homelessness. Providers may have diferent interpretations of homelessness in this shared text. This could lead to inadequacy of explaining a choice or to inconsistency in choice of a client’s service oferings across time or across service providers. Such inconsistency in choice or explanation would go counter to the ethical practice standards with which this social service organization remains compliant. For instance, linguistic Divergence leading to a consistently inconsistent choice of service oferings across service providers would be considered a source of unwanted bias, and the choices made would be unfair.
The following definition is given to the service providers to help them determine the choice of service ofering. This fictional definition is purposely inadequate to allow for illustration of the methodology:
permanent, and appropriate housing, or the means and ability of acquiring it.
In our scenario, service providers for clients experiencing homelessness make choices based on a client’s
needs. Literature diferentiates between “instrumental” needs of an organism that are associated with
instrumental goals (or, ends) and means of achieving those goals (or, agents) and “basic” (or, absolute)
needs of an organism that are inseparable from the organism itself, such as a human’s need for food,
without which a human would not exist [
In this risk-based approach, basic needs are addressed in the first step of the methodology. Specifically, an additional objective is added to the risk criteria. All four steps of the methodology are described next.
Uncertainty is a source of risk; therefore, sources of uncertainty are also sources of risk. To manage
linguistic uncertainty in shared text, each of the fourteen sources of linguistic uncertainty are addressed
following a methodology that includes steps of risk management described in the international standards
ISO 31000 [
To address clients’ basic needs described in Section 4.2, an additional objective is added. Specifically, this objective is compliance with the organization’s legal obligations and its standards of ethical practice. In our scenario, this additional objective is met by the service providers’ adherence to the organization’s policies related to clients’ basic needs. 2. Identify risk sources: Describe the sources of linguistic uncertainty that could impact the objectives defined in step 1. These are the fourteen sources of linguistic uncertainty identified in Section 2 and described in the second column of Table 1. 3. Perform risk assessment: Assess to what extent each identified source of linguistic uncertainty could lead to not meeting objectives, which in this scenario are consistency and explainability of choice of client service oferings, and adherence to the organization’s policies related to clients’ basic needs.
Risk assessment is accomplished in two steps. The first step involves estimating the impact of each of the sources of risk by posing questions to stakeholders. These questions estimate the linguistic uncertainty from each of the sources of risk (low, medium, or high) and describe the risk (i.e., the uncertainty of not achieving the desired outcome due to that source of risk).
To perform risk assessment for Complexity, the questions to stakeholders include: To what extent does applying this definition to a specific situation depend on factors not included in the definition? In other words, to what extent does this definition provide insuficient context to be able to apply it in any given situation (low, medium, or high)? In our scenario, the service providers rate this definition as having high uncertainty due to Complexity, and describe how a client’s ability to acquire housing has many dependencies such as financial ability or whether they are a dependent child. Additionally, there likely are factors about a client’s situation that are unknown to the service provider. Without acknowledging the factors taken into account in making a choice, explainability of choice of service ofering would be •To what extent does applying this definition to a specific situation depend on factors not included in the definition, or provide insuficient context to apply to any given situation? (low, medium, high)? •There are many known dependencies on a client’s ability to acquire housing such as financial ability or whether they are a dependent child. Also, likely there are factors about a client’s situation that are unknown to the service provider. Without acknowledging the factors taken into account, explainability of choice of service ofering would be lacking. Additionally, new factors may only become apparent after service provision has occurred. •To what extent can this definition be methodically interpreted (low, medium, high)? •If this definition were dificult to locate in the reference text when it was needed to make a choice, it would make it dificult to apply consistently. •To what extent is there variability across stakeholders in how the definition is applied (low, medium, high)? •Do stakeholders consistently interpret the definition in ways that would lead (consistently) to diferent choices? •How frequently might the definition need to change (low, medium, high)? •Are there foreseeable changes to policy that would require the definition to be revisited? •Is something missing from the definition (yes:high; no:low)? •If the definition were missing the word stable, it would be dificult to explain how a choice could resolve homelessness. •To what extent is it dificult to understand this definition (low, medium, high)? •For instance, do some parts of the definition translate poorly, making it dificult to understand in other languages? •To what extent might the definition be vague about something important (low, medium, high)? •If the definition did not allow the interpreter to pin down the meaning, then how to explain their choice? For instance, this definition is vague about whether permanent, afordable housing that lacks security would be considered appropriate. •To what extent must one rely on pure guesswork to interpret the meaning (low, medium, high)? •A client might require easy access to resources unknown to the client service provider such as work, school, church, or family caregivers. •To what extent might the definition obscure something important (low, medium, high)? •This definition is obscure about whether living with non-family members is a situation of homelessness, making a choice in this situation hard to explain. •Is there more than one meaning (yes:high; no:low)? •If the phrase stable housing from the definition were interpreted to mean fixed location versus reliable, this could afect the consistency of the choice. For a temporary worker, stable might best be interpreted to mean a reliable housing solution that will allow them to move as required in order to remain close to their temporary work locations. •With what frequency should the shared meaning be re-visited for a larger stakeholder audience (low, medium, high)? •Is this definition intended to be used by a new group of service providers who might interpret the meaning of the definition diferently? •To what extent are all concepts in the definition familiar enough to every stakeholder (low, medium, high)? •Not every service provider is familiar with calculating the financial means of a client, making consistency of choice across providers dificult. •To what extent is it dificult to predict how the meaning of the definition might change over time (low, medium, high)? •The meaning of appropriate housing may change in ways that cannot be easily predicted. New types of housing might be built to better accommodate temporary workers, and digital connectivity requirements for remote workers might evolve. This could afect the consistency of choices between service providers who are aware of these changes and those who are not. •Is there more than one definition that is valid (yes:high; no:low)? •In addressing the risks identified above, more than one valid definition might be warranted. For instance, diferent definitions may be helpful to making a choice depending on whether a client is unsheltered, emergency sheltered, provisionally accommodated, or at-risk.
To perform risk assessment for Unpredictability, stakeholders are asked: To what extent is it dificult to predict how the meaning of the definition might change over time (low, medium, or high)? In our scenario, the service providers consider the definition’s Unpredictability to be medium because the meaning of appropriate housing may change in ways that cannot be easily predicted. New types of housing might be built to better accommodate temporary workers, and digital connectivity requirements for remote workers might evolve. This could afect the consistency of choices between service providers who are aware of these changes and those who are not.
For Unfamiliarity, one question is, to what extent are all the concepts in the definition familiar enough to every stakeholder (low, medium, or high)? In our scenario, the service providers rate the definition’s risk from Unfamiliarity as high and explain that not every service provider is familiar with calculating means of a client, specifically financial means. This could afect the consistency of choices between service providers who are familiar with making this calculation and those who are not.
This line of questioning continues for the remaining sources of risk. A detailed list of sample questions and notes for the scenario’s risk assessment is found in Table 2.
The second step in risk assessment is to prioritize these risks according to their likelihood of impact, and addressing those that are deemed high enough priority. Highest priority is given to any risk that could result in non-adherence to legal obligations or standards of ethical practice related to basic needs. In our fictional scenario, the service organization wants to minimize any inconsistency in choice or challenge to explainability and will address all risks, making prioritization unnecessary. 4. Prescribe risk treatment: Those sources of risk that could be eliminated should be, and those remaining should be reduced as much as is feasible. There will be cases where nothing can be done about an uncertainty, and these should be identified in documentation about the decision.
In our fictional scenario, brainstorming with service providers arrives at the (sample) actions described in the third column of Table 1. These fall into 5 categories of risk treatment actions. The first four eliminate or reduce the risk to consistency of choice of service oferings: • Revising the shared text: Some uncertainties are eliminated by revisiting and improving the shared text. For example, Incomprehensibility would be entirely addressed by making the shared meaning entirely comprehensible, or understood by all Stakeholders. In our scenario, risk treatment for Complexity would include explicitly adding known dependencies, such as how the meaning of the definition should be diferent for clients who are children. • Training: Some uncertainties might require not only revising the shared text, but also providing training to the service providers on how to interpret the revised shared text. For example, Unfamiliarity could be addressed by providing training examples to help make the shared meaning more familiar to service providers. In our scenario, risk treatment for Unfamiliarity includes improved text and training on how to calculate a client’s financial means. • Calling on human judgement: Expert judgement could address tacit knowledge issues such as tacit dependencies. Human judgement reliably takes into account human attitudes that can make a decision more humane, more ethical. In our scenario, risk treatment for Complexity includes asking a client whether other considerations should be taken into account and checking in with a client after service provision to see if any new factors become apparent. • Repeating risk assessment and treatment: As previously mentioned, both Unpredictability and Impermanence might be resolved by revisiting the shared meaning of the text frequently enough for stakeholders to remain aligned on their interpretation, or, in other words, repeating steps 3 and 4 of this methodology frequently enough. In our fictional scenario, risk treatment for Unpredictability includes revisiting the shared meaning of appropriate housing frequently enough for the choices to remain consistent across service providers.
• Documenting: Some aspects of uncertainties cannot be eliminated or reduced, and must be accepted.
Such cases should be addressed by documenting the aspects of the shared meaning for which the uncertainty cannot be addressed, reasons for them, and potential impact to outcomes. For example, Insolubility may not be entirely resolvable. There may be shared beliefs, values, principles, policies and such that need to be applied to arrive at a shared meaning. Suficient details of this should be documented, as should aspects of the shared text that remain insoluble, reasons for this, and the potential impact to outcomes, which in our fictional case are consistency and explainability of choice of service oferings. Documenting interpretation allows an explanation to be more complete, and subsequent service providers could refer to this documentation to help make more consistent choices. A second example are instances where interpretation of the shared text (shared meaning) is based on tacit knowledge, which cannot be codified and requires human judgement. The details of any human judgement should be documented to describe this risk treatment. In our scenario, risk treatment for Complexity includes explicitly documenting both known factors and other factors that the client makes known.
Care should be taken so that the risk treatment for one source of uncertainty (source of risk) does not unduly add to another. One example is allowing for more than one shared meaning (adding diversity, or Variety) while not adding unwanted inconsistency, or Divergence. For example, an interpretation may be dependent on logistics of a client experiencing homelessness. Adding this dependency to the definition of homelessness would add Variety (by allowing there to be more than one valid shared meaning based on an explicit dependency). However, this diversity should be added without diverging in a way that could compromise desired outcomes, which in this scenario are the ability to consistently determine and be able to explain the choice of client service oferings, and adherence to the organization’s policies related to clients’ basic needs.
Many concepts mentioned in this section are further developed in the next, where they are referenced in an ontology of managing linguistic uncertainty in shared text.
A simplified conceptual model for a proposed Ontology of Managing Linguistic Uncertainty in shared
Text (or, MLUT ontology) is illustrated in Figure 1, which highlights in bold the MLUT components
and in italics the DOLCE taxonomy proposed to operationalize these components [
SharedText is construed as a non-agentive-physical-object (NAPO) that is part of a ReferenceDocument, which is artifact originating from a process (out of scope for this paper) of authoring the ReferenceDocument. A NAPO does not have agency; it requires an agent to become relevant within an ontology. In our fictional scenario, the agent (not illustrated in Figure 1)—an agentive-physical-object—is a natural person who plays the role of service provider or human expert—an agentic-social-object—and is participant-in a process of Interpretation.
SharedMeaning is a social-object, specifically an information-object that expresses a description (not illustrated) of the meaning of the SharedText. SharedMeaning is realized-by the process of Interpretation mentioned above and during the DecisionMaking process, which comes after an Interpretation, which precedes it.
LinguisticUncertainty is realized by SharedText (which realizes it) during the process of Interpretation. It is a collection of SourcesOfRisk. For simplicity, only one in the collection is illustrated in Figure 1, specifically Complexity. Each of the SourcesOfRisk realizes Risk. Also, each one of the SourcesOfRisk is a non-agentive-social-object, specifically a role (not illustrated in Figure 1) that is participant-in an activity of “cognitive force” occurring during the process of Interpretation. The role of the source of risk might be played-by the cognitive material of a natural-person, such as a human service provider, impacting the person’s interpretation of SharedText. More work could be done to elaborate on this concept of cognitive force. Each one of the SourcesOfRisk has-quality. In our scenario, these qualities are estimated quantitatively (as either low, medium, or high) by questions posed during risk assessment step 3 of Section 4.3.
Risk is a mental-object that is participant-in the process of RiskAssessment. Risk is imagined by agents (not illustrated) also participant-in RiskAssessment. A widely accepted definition of risk is as an efect of uncertainty on outcome; in other words, a possible efect on outcomes. This requires imagining a possible—as yet unrealized and potentially never realised—efect on outcome. More work could be done to elaborate on Risk as an imagined entity. Each Risk has-quality. In our scenario, these qualities are estimated qualitatively (in the form of a description) by questions posed during risk assessment step 3 of Section 4.3.
RiskAssessment for each of the SourcesOfRisk is a process. In this conceptualization, the RiskAssessment process includes (in a way not illustrated here) the process of Interpretation. As described above, each of the SourcesOfRisk and each Risk have qualities, and these are estimated during the RiskAssessment.
RiskTreatment for each of the SoucesOfRisk is a process that follows RiskAssessment (which precedes it). It includes more than one activity (not illustrated) that impact both (generic-target) Outcome and (generic-target) SourcesOfRisk. In our fictional scenario, there are five types of risk treatment processes: Revising the shared text, training, calling on human judgement, repeating risk assessment and treatment, and documenting. Risk treatments can both directly afect client service outcomes (e.g., documenting) and indirectly afect outcomes by modifying the risk sources (e.g., revising the shared text).
Outcome is a non-physical-endurant that has-quality that inheres-in the result (not illustrated) of the DecisionMaking and RiskTreatment processes. In our fictional scenario, risk treatments can lead to more desirable quality of outcomes, specifically improvement to consistency and explainability of choice of service oferings while adhering to the organization’s policies regarding clients’ basic needs.
This work relies on a comprehensive typology of uncertainty contextualized for linguistic uncertainty, and an approach is proposed for managing linguistic uncertainty in a fictional social services scenario where service providers must rely on a shared definition of homelessness to help them make a choice of service oferings for a client. One can imagine how it would be possible to adapt the method to other text with other groups of stakeholders making other types of decisions with other objectives. For instance, while in this illustrative scenario the shared text is a definition of homelessness, this methodology could be applied to other shared text, such as accountability found in organizational policies and procedures for AI governance.
By consciously addressing uncertainty, this methodology helps address an endogenous problem in uncertainty management, namely intolerance to uncertainty that can lead to unacknowledged bias in outcomes. Future work could be done to validate this methodology in a real-world scenario.
Future work could also explore the concept of a cognitive force of uncertainty and the risk it realizes in various scenarios. Research in uncertainty management is informed by multiple domains including risk management, behavioural decision making, ontology engineering, and others. It is hoped that this paper inspires more such collaboration.
The author thanks Michael Grüninger and Bart Gajderowicz for their guidance, and OSS 2025 reviewers for their feedback on this paper.