Fidelity in legal coding: applying legal translation frameworks to address interpretive challenges

ABSTRACT There is rapidly developing interest in legal coding, the development of machine-consumable code representations of legal rules. However, interpretive ambiguities inherent in legal rules make it challenging to directly translate law into code. Interpretive ambiguities pose challenges for legal coders, who must determine the extent to which they disambiguate rules, as well as the interpretive methods applied. Similar interpretive issues have been historically addressed in the field of legal translation. This paper argues that a legal translation framework, known as the principle of fidelity, can be used to better understand and resolve interpretive ambiguities inherent in the legal coding process. Three models of fidelity prominently discussed in legal translation literature are adapted and applied to legal coding. Accordingly, the application of fidelity models to the legal coding process could provide new understanding and methods to address complex interpretive ambiguities.


I. Introduction
[A] 'good' translation [is] not the one that represents an interpretation of the text, but the one which leaves the door open for all interpretations which the original text provides evidence for. 2 The quote by Al-Jarrah and colleagues highlights the difficult interpretive choices involved in producing legal code, the digitisation of legal rules into machine-consumable code.Legal code is increasingly necessary because it can be used for the purposes of automation and legal assistance, particularly to reduce compliance burdens stemming from complex legislation. 3Furthermore, the widespread development and deployment of legal code has been hypothesised to lead to increased transparency, efficiency, effectiveness, and fair application of legal rules. 4Consequently, legal coding development has been embraced by several governments, including in Denmark, 5 Australia, 6 and New Zealand. 7Despite the fact that there has been significant worldwide interest in legal coding, there is still substantial dispute in academia and industry regarding how to resolve the inevitable interpretive ambiguities that arise from the ambiguous nature of legislation. 8The presence of discretionary provisions has been well-documented to conflict with the deterministic nature of code. 9Similarly, linguistic ambiguities, as well as logical inconsistencies in legislation pose significant challenges for legal coders who are inherently forced to disambiguate 10 in converting natural language to deterministic code.
Early developers of legal code applications have, by default, approached legal coding in a relatively literal and isomorphic manner, assuming most statutory provisions have objectively 'correct' interpretations. 11Accordingly, whilst the legal coding literature is still young, it is nonetheless grappling with a host of unresolved challenges, such as the resolution of interpretive ambiguities.However, many of the interpretive issues faced in the production of legal code could be understood better and addressed by reference to concepts from the established field of legal translation.Al-Jarrah et al's guidance above acknowledges the inherent ambiguities present in legal texts and is therefore as potentially indispensable for legal coders, as it is for legal translators.
In contrast to the relatively nascent field of legal coding, legal translation is utilised in a variety of manners internationally and is vital to the legal systems of numerous constitutional democracies.Specific methods vary from jurisdiction to jurisdiction based on cultural needs.Switzerland, for example, has always been a multilingual state and its federal 12 legislative texts are made available in all three official languages, 13 German, French and Italian. 14Each legal translation is considered equally authentic and 'original'. 15The European Union (EU) covers 24 official languages 16 and has three 'procedural' languages, English, French and German, that are primarily utilised in internal business, 17 resulting in noticeably different drafting processes. 18Nevertheless, the EU has acknowledged the equal authority of all translated legislation. 19e argue in this paper that well-established insights from legal translation theories based on fidelity principles provide a valuable lens to consider the interpretive ambiguities arising from legal coding.Within the context of translation literature, fidelity-based approaches acknowledge the difficulty in achieving 'perfect' correspondence between all aspects of a source text and its translation, and instead propose various aspects of the source to prioritise maintenance of during translation. 20The application of fidelity principles to legal coding outlines several distinct approaches to resolve the types of interpretive ambiguity commonly found in legal coding challenges.Applying these different fidelitybased approaches will result in different types of legal coding outputs, and more importantly, will ensure that legal coders are conscious of how they resolve legal coding challenges and the limitations of their code.Section II outlines the inherent interpretive ambiguities and challenges faced by legal coders, challenging the notion that legal coding is an objective process, and supporting the need for a more nuanced approach to legal coding challenges.In Section III, the fidelity framework is outlined in the context of legal translation literature, before it is applied to legal coding challenges in Section IV.

II. Legal coding and types of interpretive ambiguity
Early attempts to convert natural-language law into code were largely based on a literal approach to interpretation, similar to historical attempts to translate law. 21Legal coding considerations have thus historically been concerned with identifying prescriptive, non-discretionary clauses, with a general belief that these clauses were relatively objective and easy to code. 22For example, one of the first attempts at encoding legislation was documented in Sergot et al.'s 1986 paper, which involved the encoding of the British Nationality Act 23 ('BNA').Their seminal paper illustrates a core tendency of legal coding research which operates on the assumption that an objectively correct interpretation of prescriptive clauses always exists. 24ore recently, optimistic proponents of legal coding similarly have implied objective or near-objective meanings can be found for most non-discretionary provisions.The Organisation for Economic Co-operation and Development ('OECD') encourages encoding 'prescriptive rules' which require 'little discretion' and 'leave little ambiguity', as opposed to rules which require subjective human interpretation. 25The notion that prescriptive rules can be effectively coded objectively is consequently one that continues to permeate throughout much of the modern legal coding movement. 26Moreover, faith in a default literal interpretation approach appears to have garnered support from industry and government.Commercially, a vast range of regulatory technology ('RegTech') offerings have developed despite a lack of solutions to identified legal coding challenges. 27This indicates that within the RegTech industry, the default literal approach to legal coding appears to be deemed 'good enough' by some for a substantial portion of relevant legal rules.The ongoing development of early governmental legal coding attempts, such as Denmark's 'digital-ready legislation', 28 as well 22 Matthew Waddington, 'Rules as Code' (2020) 37 Law Context Socio-Leg.J. 179, 182-183. 231981 (UK). 24Marek Sergot and others, 'The British Nationality Act as a Logic Program' (1986) 29 Commun.ACM 370, 371.Note that Sergot et al. did not solely adhere to a literal reading however, and considered the intended meaning of the act where a literal interpretation appears clearly incorrect.ibid 380. 25 Mohun and Roberts (n 3) 92. 26See, e.g.'The rules we can currently code effectively are prescriptiveblack or white, yes or no.': Tim de Sousa and Pia Andrews, 'When We Code the Rules on Which Our Society Runs, We Can Create Better Results and New Opportunities for the Public and Regulators, and Companies Looking to Make Compliance Easier', The Mandarin (online, 1 October 2019) <https://www.themandarin.com.au/116681-when-machines-are-coding-the-rules-on-which-our-society-runswe-get-better-results-new-opportunities-for-the-public-and-regulators-and-companies-looking-to-make-complianceeasier/>; Common themes emerged around a desire to automate and standardise planning decision making for development that is highly prescriptive and code based … statewide prescriptive development standard for low-risk developmentwas seen as a low hanging fruit for coding planning rules in NSW due to its format being highly prescriptive codes that can be interpreted into a "yes" or "no" compliance check.as domestic Australian legal code adoption, 29 seem to indicate a similar stance by government.
Legal coders also appear to rely significantly on the ability of literal approaches to produce acceptable code.For example, a majority of law and computer science students involved in a legal coding experiment conducted by Escher et al. expressed confidence in their literal legal encodings and agreed that their code sets could be used as legal advice tools for members of the public. 30Within the scholarly legal coding literature, however, many academics reject the notion of rules which can be objectively interpreted and encoded.For example, Barraclough et al. contend:   The suggestion that it is possible to [code only objective rules] anticipates that there are rules where no interpretation is required.We do not agree that any such rules in natural language exist. 31e assertion of Barraclough et al. finds support in a large body of literature concerned with ambiguity in statutes. 32It is widely accepted that ambiguity is ubiquitous in natural language on a number of levels, 33 and legal texts are no exception to this.A vast number of interpretive challenges have been identified by scholars in seemingly simple legal rules, contradicting the notion that legal coding is a straightforward and objective task, and supporting the need for a more nuanced approach.The types of interpretive ambiguity identified in the literature is now outlined, beginning with discretionary ambiguity and then on to linguistic ambiguities before concluding with logical inconsistencies.

A. Discretionary ambiguity
An obvious category of legislative ambiguity, and one that has engendered significant scrutiny in the legal coding literature, is discretionary ambiguity.Discretionary ambiguity encompasses provisions containing words such as 'reasonable' and 'likely', that grant discretionary powers to the judiciary and other parties involved in the regulatory process, to interpret what these terms mean.For example, in encoding the BNA, Sergot et al.'s noted difficulty in dealing with conditions such as 'being a good character' and 'having reasonable excuse'. 34Such discretionary provisions continue to be a regularly faced challenge in legal coding today. 35The deterministic nature of standard code means that it is difficult for legal coders to convert discretionary clauses with objective certainty 36 due to their widely recognised subjective nature. 37Accordingly, in order to fully encode discretionary provisions in a functional and meaningful manner, legal coders must disambiguate the provisions.
Disambiguation essentially means that the coder is required to apply their own discretion and interpretation. 38Further, the coder's discretion and interpretation are built into the code, and therefore applied to all users of the code.Needless to say, this process gives rise to a number of key issues, not least, the power to definitively disambiguate the law is one that is constitutionally exclusive to the judiciary and is a generally accepted rule of law principle. 39Hence, any legal code seeking to disambiguate discretionary clauses cannot be said with certainty to produce legally determinative outputs that are legally accurate to a strong degree.The legal coder can obviate these challenges by leaving evaluation of discretionary provisions up to a human user's interpretation, 40 for example, the commercial entity that the code has been produced for, but this will significantly reduce the utility and functionality of the legal code, especially in RegTech systems used for digital compliance purposes.As such, it is not possible to avoid interpretive challenges arising from discretionary clauses in legal coding exercises, despite the historical and contemporary contentions portrayed in the literature.The same can equally be said for linguistic ambiguities.

B. Linguistic ambiguities
In contrast to discretion, which is clearly visible to the legal coder, ambiguities of a linguistic nature are much less obvious and hence pose a greater risk to the accuracy of legal code produced by literal conversion methods.Linguistic ambiguities can take a number of forms, such as syntactic, semantic, anaphoric, pragmatic, and more. 41A syntactic ambiguity is a form of lexical ambiguity which refers to an ambiguity arising from the categorisation of a given word, each reading of which will generate different understandings of the broader text.For example, the word 'pen' might refer to a noun (such as a writing implement or an animal enclosure) or a verb (to write something).Syntactic ambiguities can often (but not always) be resolved through contextual clues. 42A semantic ambiguity, like a syntactic ambiguity, is also a form of a lexical ambiguity.Semantic ambiguities, however, refer to ambiguities which are not syntactically ambiguous, but still have multiple meanings arising from a given word.Continuing the example of the 'pen', the phrase 'I plan to sell my pen' might generate semantic ambiguity if one owns both an animal enclosure and a writing implement as both are nouns which could be sold, but no longer generates the syntactic ambiguity arising from the verbal meaning, 'to write'. 43naphoric ambiguity refers to ambiguities which arise due to multiple antecedents occurring in a sentence prior to a pronoun.As a result, readers can be confused as to which antecedent the pronoun refers. 44For example, in the statement -'The translator must convert the text from the language of the seller into the language of the buyer, and they must be satisfied with the translation'the use of 'they' could refer to the translator, the seller, or the buyer.Finally, a pragmatic ambiguity is arguably a more problematic ambiguity than its syntactic or semantic counterpart, referring to an ambiguity in which a sole meaning can be applied in multiple ways in one context. 45For example, in a text which defines and regularly refers to two separate individuals, the phrase, 'the individual is obligated to do x' is pragmatically ambiguous, as it is unclear which individual is being referred to, and cannot be disambiguated by reference to the wider context of the text.Some linguistic ambiguities are less likely to appear in legislative texts given the limited contextual scope and semi-formalistic manner in which legislation is drafted. 46evertheless, even seemingly simple provisions are often afflicted with some form of linguistic ambiguity. 47Ashley, for example, outlines the various forms in which ambiguity can present itself in a legal text, identifying semantic ambiguity, vagueness, and syntactic ambiguity as common occurrences throughout statutory texts. 48He shows that even where legislative extracts lack obvious or intentional ambiguity, 49 unintentional polysemy 50 necessitates interpretation on the coder's behalf.Allen & Saxon's seminal study 51 is used to highlight that 48 interpretations of varying strength were generated out of a two-sentence long provision due to syntactic ambiguity. 52Engholm similarly provides examples in which syntactic ambiguity leads to multiple valid readings of provisions, including where a particular prohibition can be read as either a 'condition, a conditional consequence, or an unconditional consequence'. 53It should therefore come as no surprise that the same challenges persist in contemporary legal coding exercises.Witt et al., for example, demonstrated that legal coders who are similarly trained and are working collaboratively, can encode the same legislative texts in significantly different ways. 54t also should be noted that linguistic ambiguities are often deliberately inserted into law by lawmakers through a practice known as 'compromise-focussed ambiguity'. 55In these cases, vagueness and semantic ambiguity are utilised in the drafting and revising phase of lawmaking to reach political compromise when multiple parties cannot agree on what a legislative clause should state. 56While the root cause of ambiguity here differs from the unintentional polysemy discussed above, the methods by which the ambiguity emerges are the same.Perhaps unsurprisingly, this phenomenon also arises in the field of legal translation, where it is referred to as 'strategic ambiguity'. 57he existence of intentional ambiguity raises the question of whether legal coders should in fact even attempt to disambiguate clauses, just as legal drafters have been known to deliberately leave ambiguities in legislation because they lack the authority to disambiguate. 58he above highlights the challenges for legal coders in resolving interpretive challenges based on linguistic ambiguities.These ambiguities are often unintentional, but they can also be the result of political compromise, which coders should avoid disambiguating to the extent possible.This point is reemphasised in relation to the issues that arise from logical inconsistencies in legislation.

C. Logical inconsistencies
Interpretive ambiguity challenges can also relate to potential logical inconsistencies in normative texts such as statute.Barnes presents 59 a straightforward example of logically inconsistent statutory language in the case of Smoker v Pharmacy Restructuring Authority 60 ('Smoker'), where disagreement arose surrounding the interpretation of Section 99K(2) of the National Health Act 1953 (Cth) 61 ('the NHA'), which states: In making a recommendation … the Authority must comply with the relevant guidelines determined by the Minister under section 99L. 62he inconsistency in the Smoker case arose due to an apparent contradiction between the word 'must', used to imply a mandatory obligation, 63 and the word 'guidelines', which typically implies non-compulsory recommendations. 64Ultimately, while it was held that the creation of a mandatory obligation was the intent of the legislation, 65 the court acknowledged that a drafting inconsistency was present and was only able to resolve it through a wider contextual reading of the legislation. 66ogical inconsistencies can also take additional forms, such as the existence of two provisions which contradict each other.Despite their time-worn existence, there is no methodological means to resolve logical inconsistencies present in the legal coding literature.The de facto resolution process, based on the default literal approach to legal coding would see programmers simply coding both contradictory provisions. 67This might be a practical means to conduct a literal conversion of natural language to code, in as objective a manner as possible, but it reduces the predictability of the law and increases the risk of arbitrary decision-making which is antithetical to the rule of law, 68 and undesirable in industry use-cases.
It is possible to resolve logical inconsistencies in legal coding by moving beyond the realm of the literal language of statutory texts.Priest notes that in some cases, statutory inconsistencies can be resolved by reference to 'implicit exceptive clauses'. 69He suggests that appeals to superior laws, 70 considering the stated purpose of the statutory text, 71 or applying the principle of lex posterior 72 can resolve these 'implicit' clauses. 73Priest notes, however, that these implicit exceptive clauses cannot clarify all inconsistencies. 74Hill approaches this challenge from the perspective of conflicts between Australian state laws. 75He shows that some seemingly inconsistent laws often have pathways to resolution, but there are also scenarios in which two laws may be genuinely inconsistent. 76Applying the approach of the Australian judiciary, the High Court case of Project Blue Sky Inc v Australian Broadcasting Authority 77 ('Project Blue Sky') largely defines the modern approach to statutory interpretation. 78he High Court noted that where a set of provisions conflict with one-another, the conflict must be alleviated in a manner which best maintains the overall purpose of the statute. 79his process often will involve determining the hierarchy of the conflicting provisions and will resolve most logical inconsistencies. 80he arguments presented above outline challenges that arise in the production of legal code.Employing the process of Project Blue Sky may be a reasonable task for a member of the judiciary, but it is questionable whether it is one which should be expected to be undertaken definitively by legal coders. 81Such a task requires coders to definitively determine whether or not a contradiction exists, identify what, if any, hierarchical relationship exists between the relevant provisions, and encode an outcome which aligns with the modern approach to statutory interpretation.It is not difficult to imagine how such a scenario would likely see numerous functionally different encodings, particularly where a hierarchical relationship is absent or difficult to identify.A more optimistic perspective, however, would note that administrators are generally expected to emulate the judiciary's approach to statutory ambiguities, 82 and therefore placing this expectation on legal coders may not be unrealistic.Nevertheless, the ideal approach of legal coders to statutory inconsistencies remains an open question, suggesting an additional difficulty in legal coding exercises that seek a straightforward one-to-one 'translation'.
Building on the resolution of interpretive ambiguities highlighted above, the next section introduces fidelity theories of legal translation to understand the interpretive contexts of legal coding in a more sophisticated way.

III. Legal translation theories and fidelity models
Section III outlines fidelity principles from the legal translation literature that could be employed as new approaches to better understand and address interpretive ambiguities in legal coding exercises.Before we get that far, however, it is necessary to deal with a potential criticism of the fidelity model approach.As noted above, legal coding research has historically and recently been focused on 'translating' the law through a one-to-one encoding exercise.A comparison between narrower legal coding interpretive approaches and broader legal translation concepts could therefore appear problematic.In other words, the legal coding process does not involve a translation process that would be readily understood in the legal translation literature.This understanding of 'translation' has led to criticism surrounding comparisons of translation and coding from legal scholars who oppose the literal approach to legal coding. 83We counter this potential criticism by arguing that legal coding research, particularly in the relation to proponents of the Rules as Code movement, is built upon an outdated notion of translation.In fact, the contrary is the case.Modern translation methods are appropriate guides to legal coding, because they provide a more sophisticated notion of translation that goes beyond a one-to-one exercise.
Moreover, specialised translation fields, 84 such as legal translation, have already addressed many of the interpretive challenges that are currently being faced by legal coders.Legal translation now has its own field 85 which has developed methodological approaches that acknowledge the unique requirements of translating legislation.While legal coding scholars grapple with questions of authority, logistics, and equivalence in developing and publishing encoded legislation, legal translators have historically raised similar questions regarding their translated provisions.For example, just as ambiguity proves to be a significant challenge for coders, it is also a problem for legal translators, who have to work with texts featuring unintentional ambiguity, deliberate ambiguity for the purposes of discretionary power, and strategic ambiguity.
Modern legal translation approaches identify three key concerns: equality of language, efficiency of translation, and fidelity in interpretation. 86Section III focuses on fidelity principles because this concept has most relevance to resolving the interpretive ambiguities alive in legal coding.The maintenance of fidelity in translation is considered one of the greatest sources of debate in translation theory. 87Upholding fidelity involves translating the source text as a whole, rather than simply translating each word in the source text on a one-to-one basis.Legal translation, like legal coding, has historically been conducted in a strict literal sense.However, over time, the legal translation field, unlike its legal coding counterpart, has developed more nuanced translation approaches which increasingly value maintaining fidelity to the source text. 88n early example of legal translation's interest in fidelity principles is Rossel's nonliteral translation of the Swiss Civil Code 89 in 1907.Rossel's translation from German to French was heavily criticised 90 but it started the critique of default translatory approaches as one-to-one translations.Rossel argued in favour of a 'natural' translation which emphasised clearly communicating the text's meaning to the French-speaking population of Switzerland over preserving the exact syntax and terminology. 91Rossel's rejection of translational literalism, the theory that any text canand, according to literalists, shouldbe translated to the literal word, with no interpretation on the part of the translator, was the first of many throughout the century which served as foundations to the modern approaches to legal translation.
In the second half of the twentieth century, translation scholars started to increasingly embrace the notion that translation was more than a literal reproduction of the original 84 Specialised translation refers to the translation of 'specialised' texts which require special knowledge outside of a typical translator's expertise, such as legal texts, scientific texts, religious texts, etc. 85  Law 277, 279-280. 87Charles-Green (n 21) 183. 88See, e.g.Šarčević, New Approach To Legal Translation (n 20) 23-53 for a historical review of legal translation. 89Schweizerisches Zivilgesetzbuch [Swiss Civil Code] (Switzerland). 90Šarčević, New Approach To Legal Translation (n 20) 37-40; Charles-Green (n 21) 183. 91Šarčević, New Approach To Legal Translation (n 20) 36-40; Charles-Green (n 21) 183-184.
text in a new language. 92In the legal translation field, multilingual jurisdictions began to follow suit.Canada, for example, started to treat legal texts in both official languages, English and French, as equally authentic. 93The Canadian approach permitted the translator, now considered a co-drafter, to re-word and delete text in translating the source legislation, so long as fidelity was preserved. 94This approach has been considered widely successful in the eyes of Canada's Parliament and Supreme Court as well as the wider linguistics community. 95In light of such reforms, many linguistic scholars now accept that fidelity is concerned with maintaining an 'equivalent impact on the target reader, which may justify substantial changes to the original text'. 96teps such as Canada's treatment of the translator as a co-drafter and the EU's handling of translated texts as separate pieces of legislation reflect the different theories of translation, and legal translation that subsequently developed.For example, Mannoni elaborated on Popovič's theory of prototexts and metatexts, 97 to argue that the translation product of a given source text (the 'prototext') is an entirely new text whose sole author is the translator (the 'metatext'). 98He illustrated several translations of a clause deploying various levels of literalism and non-literalism, and found that only non-literal translations accurately represent the text. 99The idea that legal translation regards the production of a new metatext is now common in the modern legal linguistics literature. 100hile there is a consensus in the literature that legal translation should be concerned with ensuring that a given metatext maintains fidelity to its prototext, there is less consensus regarding the attributes of the prototext to which fidelity ought to be owed.Different models of fidelity exist which vary according to how source-oriented or receiver-oriented they might be.Source-oriented models resemble literal and formalistic approaches which prioritise the source text itself, while more receive-oriented models are arguably more flexible and better align with approaches that are not necessarily literal but are instead concerned with how an individual understands the translation. 101his paper considers three of the most prominently discussed fidelity models in legal translation: Fidelity to Text; Fidelity to Legal Instrument, and Fidelity to Function.

A. Fidelity to text
Translators who lean towards literalism are most likely to embrace Fidelity to Text, which reflects the literal conversion approach.Historically, a strict literal translation constituted 'the golden rule', particularly in the context of legislative translation, 102 which often meant adhering to a word-for-word equivalence. 103Modern advocates of textual fidelity acknowledge the difficulties of attaining word-for-word equivalence, and argue that translators should strive for 'a semantically and syntactically literal translation'. 104iven literal translation's longstanding history as the de facto interpretive approach, it should come as no surprise that the approach exists in various forms, each interpreting 'literal' in slightly different manners.Baker, for example, outlines how a translator might achieve equivalence at a word level, 105 a lexical level, 106 a grammatical level, 107 and textual levels. 108Each level represents decreasingly strict levels of 'literal' translation with textual fidelity due to the emphasis on word-for-word equivalence.
Historically, from a general linguistics and information theory perspective, advocates operated on the understanding that a translator's task was one of encoding a sender's message so that it can be understood by a receiver. 109Consequently, the 'message' is taken to be constituted solely by the text itself.In other words, the text speaks for itself and does not require the interpretation of a translator to elucidate meaning.Unsurprisingly, early constructions of textual fidelity were founded on the formalistic construction and understanding of legal and religious rules. 110The same type of textually literal argument has purveyed in the legal translation field.Paraphrasing or altering the syntax of legislation inherently pushed the translator into the role of interpretation, a role constitutionally reserved for the judiciary. 111he omission of a single word in translation, regardless of its substance, thus constituted an alteration of the source text, and infringed upon the duties of legislative drafters. 112ome Fidelity to Text scholars argue that the level of literalness required in a translation is dependent upon the metatext itself.There is a foundational difference between primary legal texts, such as legislation, and secondary texts such as books or other academic works.Weisflog, for example, disapproves of non-literal translation when dealing with primary legal documents that create law, such as legislation, treaties, and regulations. 113However, a non-literal approach is acceptable to capture the author's 'thoughts and ideas' when translating secondary materials. 114Similarly, Didier advocates for a strictly literal translation of normative legal texts, but accepts a non-literal translation of judgements. 115Accordingly, the degree of strictness applied to a fidelity-based literal translation can vary depending upon the legal material being translated.
Similarly, in a general translation context, Fidelity to Text proponents, such as Hatim and Mason, contend different levels of textual fidelity are necessary based on attributes of the source text's elements. 116A textual element can be placed along a continuum stemming from 'stable' to 'dynamic'.The former refers to elements in which reader expectations are met and language norms are consistently applied.The latter refers to reader expectations that are subverted and norms which are 'systematically departed from'. 117Stable elements are more suited to a literal approach. 118For example, a news report, which is likely to be of an 'expository' nature, is arguably appropriate for literal translation, while an argumentative editorial, which contains more complex structure and opaque texture would not be. 119Hatim and Mason also identify legal texts as expository or instructional, 120 and therefore stable and suitable for literal translation.This sentiment arguably aligns well with the formalised nature of legal drafting, 121 and suggests that legislation should require literal forms of textual fidelity translation.
Applying these arguments to legal coding, it could be argued that textual fidelity is an appropriate approach for coders due to the stable and authoritative nature of legislation.However, prioritising fidelity to the literal text of legislation is unlikely to adequately address the interpretive ambiguities outlined in Section II.With this in mind, we consider more contextual and receiver-oriented fidelity models.

B. Fidelity to legal instrument
The Fidelity to Legal Instrument approach rejects the literalist approach of the Fidelity to Text model.Rather than providing fidelity to textual equivalence, translators must 'attempt to preserve the unity of the single instrument by producing a text that expresses the uniform intent'. 122In the case of legislation, therefore, the translator's primary concern should be about capturing and expressing the intent of lawmakers to ensure the intended effect of the law is maintained.To do so requires interpretation and translation of legislative preambles or purpose clauses to identify the core expressions of the legal instrument's intent. 123The goal of the translator in a Fidelity to Legal Instrument approach is to produce parallel texts that could be uniformly applied and interpreted by courts. 124idelity to Legal Instrument represents the type of 'contextual and purposive approach' which reflects the modern foundation of statutory interpretation, but in a legal translation context.The degree to which the modern approach is intended to be applied by translators surfaces some of the key critiques of Fidelity to Legal Instrument in the legal translation field.For example, the ability of a translator to accurately predict how the courts would interpret and apply a legal instrument prototext is questioned, particularly given that judges disagree amongst themselves in interpreting and applying the law to any given case. 125Moreover, it is questionable whether anyone, never mind a translator, can find the true and objective 'legal intent'. 126The translator must maintain not just translation skills, but also a significant level of legal expertise to realise the interpretation of legal intent. 127Equivalently in the legal coding context, a legal coder would require a high level of legal expertise combined with significant programming proficiency.
Despite its greater skillset requirements, the Fidelity to Legal Instrument approach is a leading method in modern legal translation, and its alignment with the modern approach to statutory interpretation makes it a promising candidate for legal coding.However, the Fidelity to Legal Instrument approach is primarily built upon constructing a metatext which carries equivalent legal effect to the prototext, and is therefore designed to produce translations with identical function.128This is the case from one translation of natural-language legislation to another, where both texts serve to declare the law, but is not necessarily the case in the context of legal coding.In the latter situation, the machine-consumable code could serve several functions other than merely declaring the law.Hence, within a legal coding context, it is important consider fidelity models which account for a change in function between prototext and metatext,129 such as Fidelity to Function.

C. Fidelity to function
The most receiver-oriented model considered is Fidelity to Function.The Fidelity to Text and Legal Instrument models consider application of literal or non-literal translation on the function of the prototext. 130Functional translation theories, such as Fidelity to Function, prioritise the purpose and function of the metatext over all else. 131Vermeer's Skopostheorie, 132 is the most notable of the functionalist theories.The translator's role is to maintain the 'skopos', or purpose, of the metatext. 133The skopos, in this regard, is a function of the needs of the 'client' that commissions the translation. 134The translator must therefore work towards the purposes set out in the client's translation brief, which may be tangible or implied, and which should be a function of the prototext's target audience. 135Advocates of Skopostheorie, such as Nord and Reiß describe the theory as both pragmatic, in the sense that it prioritises outcomes of translation, 136 and comprehensive, in that it generally is applicable to any text genres, including legal texts. 137However, not all commentators agree.
Critics of Skopostheorie argue that the functional approach is inappropriate for authoritative texts such as legislation.Accordingly, it is less appropriate for specialised translation tasks such as legal translation. 138In such specialised translation contexts, the skopos of the metatext tends to match the prototext which renders the primary motivations of Skopostheorie irrelevant. 139However, as noted above, legal coding differs from legal translation when considering the function of the metatext.Typically, the function of legal code is not simply to declare the law, but to apply it to sets of facts or processes, and produce outputs based on that application.Hence, in the context of legal coding, the Fidelity to Function approach is perhaps more applicable to legislation than in the context of legal translation.
The different types of fidelity model encountered in the legal translation field could apply to the legal coding context.The application of the models to legal coding exercises would impose different interpretive values and requirements on the coder.The interpretive role of the legal coder would therefore be substantially different in the sourceoriented models of Fidelity to Text and Legal Instrument and the receiver-oriented model of Fidelity to Function. 140Understanding the role that each fidelity model brings to the legal translation process, and could bring to the legal coding exercise, provides much need insight into legal coding activities from the perspective of literal source-oriented and non-literal receiver-oriented spectrums.The application of different models helps to explain the interpretive approach applied and its relevance to broader and contemporary understanding of legal translation, which in itself critiques the application of interpretive literalism's de facto position.The application of fidelity models also provides stronger justifications regarding the reliability of legal code outputs by being clearer on the interpretive basis in which ambiguities were identified and resolved.Consequently, if we are to adopt the fidelity framework from legal translation literature, we must also carefully consider how each model could inform and enhance the legal coding process and its outputs.Section IV considers both.

IV. Using fidelity models to address interpretive ambiguities in legal coding
Section III highlighted that modern legal translation methodologies are no longer wedded to a literal or one-to-one conversion from one language to another.Contemporary approaches involve the creation of a new metatext in a different language which maintains fidelity to the original prototext.The application of fidelity principles to translation processes offers a nuanced set of approaches which acknowledge and integrate the interpretive nature of translating a text.We now contend below that applying the fidelity models from legal translation to legal coding can provide new methods to understand and address interpretive ambiguities that will inevitably arise.The three fidelity models are now applied to legal coding practice to outline four core components of encoding law exercises.In doing so, we highlight the different objectives that arise from the application of each model which provides a better understanding of the types of interpretive method to be applied under each model.We then look at the different types of encoded output that each model produces which are based on different isomorphic considerations.It should also be noted that the four different components, detailed in Table 1, are not distinct, and crossover in application.That said, there is value in delineating them separately in order to demonstrate how different fidelity models would apply to legal coding exercises.

A. Fidelity to text legal coding
Applying the four components of the Fidelity to Text model, we can better see the often-implicit assumptions of legal coding exercises based on a literal interpretive approach.A legal coder subscribing to this model would be expected to promote objectivity to the highest extent possible to avoid building their own interpretation of law into the code.The goal of a Fidelity to Text coder is solely to produce a machine-consumable form of the source text as a standalone document.That is, to the extent possible, the language and structure of the code should resemble the natural-language legislation, such that the code can be said to be a true representation of the legislation as a standalone document. 141Consequently, an individual reading the natural-language text without any additional context or legal proficiency should come to the same conclusions that they would reading the code.From an isomorphism perspective, the Fidelity to Text approach strives for structural and rule-based isomorphism, whereby each legal unit142 is consistently separated into its own storage unit, 143 and each legal rule is represented by a specific chunk of code, and vice versa.Given that the Fidelity to Text coder strives to encode a faithful representation of the source text in a vacuum, the model inherently demands that the coder to consider no other resources throughout the coding exercise, including in the resolution of interpretive ambiguities.A Fidelity to Text coder thus only codes what is on the page before them and it is not their role to identify legal meaning through consideration of case law, or other means of legal explanation.
A question then arises based on Baker's levels of literalness in translation outlined above, as to what degree of strictness a Fidelity to Text legal coder ought to adhere to.This question is arguably simple to answer by analogy to translation theory.It is noted that a lack of equivalence in terminology between source and text language can cause difficulties in lexical-based translation. 144For example, El-Farahaty identifies difficulties in lexically equivalent translation between Arabic and English, belonging to Semitic and Indo-European lingual families respectively, as an example of such inequivalences. 145n reference to dealing with Arabic-English terminological inequivalences, Baker suggests strategies such as paraphrasing and omission,146 both of which violate the strictest historical understandings of literal translation.
If the terminological inequivalences between Semitic and Indo-European languages are apparent enough to force the translator to abandon a strict word-for-word understanding of textual fidelity, it is likely that the same could be true when translating between natural and programming languages.Thus, if legal coders are to adopt a literal approach which aligns with textual fidelity, it would likely be one which is not so strict as to prioritise word-for-word equivalency, but rather, more lenient lexical or grammatical equivalencies.In fact, the legal coding literature is replete with these types 'lenient' textual fidelity research outputs.The default approach to legal coding has been one of isomorphic literal conversion which inherently aligns with textual fidelity, but which, by nature of code syntax, cannot achieve word-for-word equivalency. 147n example of how the Fidelity to Text legal coder would attempt to resolve interpretive ambiguities can be seen in the handling of discretionary provisions in the legal coding experiment conducted by Escher et al.The researchers found that computer science students tended to code discretionary provisions as Boolean variables whose values could only be set by initial human input. 148That is, computationally, the question of whether something was 'reasonable' was interpreted to be a binary value which is determined solely by the user of the legal code application, rather than any underlying computational logic.While such an interpretive approach arguably best matches the actual semantical meaning of a discretionary provision, it ultimately places the onus on the user of the legal code to pre-empt how the judiciary would resolve a given ambiguity without explicitly signalling to the user that such discretionary questions must be approached according to particular legal rules.From a purely computational and literal perspective, a variable representing a discretionary finding is treated and presented no differently to a variable representing an objective fact.Hence, while this approach maintains fidelity to the text as a standalone document, it is limited in its ability to adequately convey the meaning and significance of discretionary provisions and is thereby vulnerable to similar criticisms faced by its counterpart in legal translation contexts.
Just as the Fidelity to Text coder sidesteps any interpretation of discretionary provisions, they also are under no expectation to acknowledge or resolve other interpretive ambiguities such as linguistic ambiguities and logical inconsistencies.The Fidelity to Text model provides no guidance towards identifying or resolving these challenges, and instead would encourage coders to, to the extent possible, bake the ambiguities into the code, passing the task of disambiguation onto users of the code.
For example, returning to the case of Smoker, a Fidelity to Text coder might simply include both 'must' and 'guidelines' in the name of the legal conclusion which arises when the Pharmacy Restructuring Authority is making a recommendation.In doing so, the legal coder can maintain the wording of the NHA, without encoding any deontic logic which imposes that 'must' creates an obligation, or 'guideline' creates non-compulsory advice.Consequently, a user of the code would still face the same ambiguity which would have arisen had they consulted the natural-language legislation, leaving the interpretation of the provision out of the hands of the coder.
Such an approach has two strengths, in that it reduces the level of legal expertise required of a legal coder and leaves the interpretation of any ambiguities to the enduser just as natural-language legislation ambiguities must be addressed by the reader.However, simply sidestepping these ambiguities also gives rise to several limitations of code produced under this model.The building of ambiguities and inconsistencies into the code severely reduces the functionality of the code, in the sense that users of the code who lack legal training are likely to get minimal benefit from it that they would not have also gotten from the natural-language legislation.Furthermore, not all ambiguities can be fully sidestepped in the manner illustrated in the Smoker example.This would be the case when considering linguistic ambiguities which demand disambiguation, or which are so subtle that the coder is not aware of possible competing interpretations.
Accordingly, the Fidelity to Text model offers a method to minimise interpretation on the part of the coder, but results in code with reduced functionality and benefit for the end-user.Furthermore, while the model decreases the required legal skills of the coder, it also offers no guidance towards identifying and addressing interpretive ambiguities which sometimes cannot be avoided.

B. Fidelity to legal instrument legal coding
Where coders of the textual fidelity approach would be expected to consider their legislative text in a vacuum, the Fidelity to Legal Instrument model emphasises interpreting and encoding the text within its broader legal context.Accordingly, unlike a Fidelity to Text coder, who could easily be the computer science students highlighted above, the Fidelity to Legal Instrument coder would require legal expertise.Fidelity to Text code should be understandable as identical to natural-language text regardless of a reader's legal proficiency.However, the Legal Instrument code will require a degree of legal understanding to recognise alignment with judicially defined processes of interpretation and understanding of the natural-language law.As noted in Section III, this would, in most contemporary approaches, stem from a contextual and purposive reading of the law, as informed by the modern approach to statutory interpretation.
Following the modern approach, the Fidelity to Legal Instrument model refers to the context provided by the broader legislative instrument, including extrinsic materials such as explanatory memoranda to fully ascertain and convey the legal instrument's intent in the code.In keeping with the objective of maintaining the legal instrument, Fidelity to Legal Instrument coders should also refer to common law, though caution should be exercised in doing so.In addition to the unreliability of the application of stare decisis by the courts, the encoding of a common law legal rule, which by its nature is not present in the source legislative text, arguably results in the inclusion of rules which are auxiliary to the primary legal instrument, and therefore alters the legal effect of the source text.To this end, a Fidelity to Legal Instrument coder should arguably avoid encoding anything other than the binding aspects of a judgement, the ratio decidendi, and should only do so where the ratio assists in statutory interpretation, as opposed to tangentially adding to the law contained in the statute.
With regard to isomorphism, the Fidelity to Legal Instrument model suggests that despite referring to extrinsic materials, coders should be careful to avoid overly reformulating or restructuring the prototext, the coded output. 149Avoidance in this manner fosters transparency and allows a reader of the code to both isomorphically map coded rules to their natural-language counterparts and understand the intent and intended effect of the legal instrument. 150However, just as Šarčević encourages minor sentence-level liberties in translating a text in order to better convey the legal intent and effect, Fidelity to Legal Instrument coders should not refrain from departing from total isomorphism where doing so better conveys the legal intent to users.For example, a Fidelity to Legal Instrument coder might include in-code annotations, such as in-line comments or flags, or other non-substantive outputs which clarify unintended ambiguities stemming from the code itself and emphasise legally important concepts.Huggins et al., for instance, encoded discretionary terms as Boolean questions to be answered by the user. 151In contrast to Escher et al.'s participants, however, Huggins et al. explicitly highlighted the discretionary nature of relevant variables to users through use of flags in the code. 152Through this approach, the discretionary elements are still computationally treated as Boolean values, but the legal significance of the variables and the requirement of human interpretation of the law is accurately emphasised to users.Hence, the approach of Huggins et al. towards discretionary provisions could largely be seen to align with the Fidelity to Legal Instrument model.
The Fidelity to Legal Instrument model best aligns with the modern approach to statutory interpretation, suggesting a higher level of legal accuracy than other models, but also demanding a higher level of legal expertise of its coders.While the model is more receiver-oriented than the Fidelity to Text model, in that it emphasises communicating the legal intent accurately, it still does not consider the actual use-case(s) of the code, and therefore cannot be tailored towards optimal functionality.

C. Fidelity to function legal coding
As the most 'receiver-oriented' in orientation, the Fidelity to Function model inherently demands that the legal coder is guided at all times by the client brief, which outlines the purpose of the code.It is possible that the final Fidelity to Function code could have a largely separate meaning compared to that conveyed by the natural-language text, depending on the functional purpose outlined in the brief.This would be the case, for example, in a compliance context, in which only rules that are relevant to the compliance-obligated entity are encoded, and those rules are encoded in a manner which only addresses the specific workflow processes of that entity.In that regard, the methods employed by a Fidelity to Function coder, and the ultimate coded output, do not necessarily have the same isomorphic, objectivity or legal alignment requirements of the other two fidelity models.Understanding of the coded output is thus purely centred on the functional utility of the code, as defined by the brief, rather than the logical dictates of literalism or the legal expectations of statutory interpretation.In that sense alone, the 'receiver-oriented' basis of the Fidelity to Function model is foundationally different to the 'source-oriented' basis of the other models.
Unlike the Fidelity to Legal Instrument's inclusion of binding extrinsic materials, or the Fidelity to Text's refusal to consider, the Fidelity to Function approach allows for the incorporation of any texts that are necessary to fulfil a coding brief.Depending on the purpose of the code, this might include regulatory materials, such as regulatory guides, industry reports, scholarly articles, or even binding materials of other jurisdictions, if the coder believes they might be indicative of how a rule should be interpreted and encoded in the context of a brief.The benefit of the Fidelity to Function approach in interpretation can be illustrated by returning once more to the case of Smoker, in which the compulsory nature of a set of 'guidelines' was questioned.
The court noted that the root of the drafting inconsistency can be tracked back to an industry agreement prior to the drafting of the NHA between the Minister and the Pharmacy Guild of Australia,153 in which it was agreed that, 'subject to the guidelines … not being disallowed by the Federal Parliament, the Authority will consider applications for approval based on the following guidelines'. 154By additional reference to the second reading of the amendment responsible for the inconsistent provision, the Community Services and Health Legislation Amendment Bill 1990, 155 Barnes further shows that an objective of the Bill was to give effect to the agreement. 156Hence, through reference to nonbinding materials, a legal coder adopting the Fidelity to Function model would likely have successfully identified the judicially-deemed correct interpretation of s99K(2) whereas a Fidelity to Text or Fidelity to Legal Instrument counterpart, may not.
As with all aspects of the model, the methodology of the Fidelity to Function approach is entirely dependent upon the brief.To visualise an approach which might suit the Fidelity to Function model, we can return once more to the encoding of discretionary provisions observed by Escher et al.In contrast to coding teams comprised of solely computer science students, the researchers found that 'mixed' teams consisting of computer science and law students tended to incorporate tests applied by the judiciary into their encoding.The teams either utilised the same language as judicial decisions or encoded lists of factors the judiciary utilised in prominent relevant cases. 157The approach adopted moves well beyond merely coding a given legislative text.Further, while this approach admirably attempts to align itself to the judiciary's method of approaching disputes, it assumes a total adherence to stare decisis, the notion that courts will follow precedent, and one which does not always hold. 158Thus, the incorporation of non-ratio aspects of a judgement, such as obiter dicta, 159 has been argued to increase the risk of an incorrect conclusion 160 compared to simply placing the interpretive onus on the user.Hence, this type of approach would be unsuitable for Fidelity to Text or Legal Instrument coders.
However, this approach might be ideal for Fidelity to Function coders whose skopos briefs involve risk reduction.In such a scenario, the purpose of the code is not necessarily intended to be an accurate representation of the law.Instead, the purpose of the code is to implement a business logic which reduces the likelihood of a client breaching the law whilst acknowledging that there might be different risk appetites arising from a regulatee's workflows.Therefore, encoding common law factors in a manner which allows for discretionary provisions to be considered satisfied but errs on the side of caution could be an appropriate methodology to apply in response to such a brief.For example, the Australian Law Reform Commission ('ALRC') notes that the Corporations Act's definition of 'financial product' is 'over-inclusive', and potentially indicates policy intent to err on the side of caution and protect consumers. 161This would suggest that a Fidelity to Function coder producing code on the behalf of a regulator dealing with financial products, such as a ASIC, might benefit from overestimating the application of compliance burdens, rather than underestimating, in cases of ambiguity.
As noted in Section III, while the Fidelity to Function method is employed in various processes of general modern translation, it has largely been frowned upon by legal translators as a tool for legislative translation, due to the specialised purpose of legislative texts. 162owever, unlike the translation of legislation from natural-language 'A' to natural-language 'B', where the purpose is invariant, 163 the digitisation of a piece of legislation involves the creation of a metatext whose purpose differs from that of its prototext.Indeed, acknowledging that nobody's primary language is a coding language, and therefore nobody would reasonably produce legal code from the primary purpose of being read by a human as a substitute for natural-language law, we are forced to accept that legal code inherently serves alternative purposes to natural-language law, 164 such as automation of workflow processes and drawing specific legal conclusions about specific sets of facts.This key difference between legal code and translated law thus circumvents the criticism that Skopostheorie is not designed for translations 'without a shift in function', 165 which otherwise applies to legislative translations.Legal coding may therefore be uniquely suited to a Fidelity to Function approach in a way in which traditional legislative translation is not.

V. Conclusion
Despite increased interest in legal coding both commercially and in government, there is no existing consensus regarding a methodology which addresses the types of interpretive challenges that legal coders face.This paper has outlined interpretive ambiguities which can arise during the legal coding process and shown that these same challenges are present within the field of legal translation.Through application of three of the primary models of fidelity considered in modern legal translation literature, the paper outlined three potential approaches to legal coding.
The Fidelity to Text approach arguably avoids subjectivity to the greatest extent.However, its inability to consider context or purpose renders it less effective in identifying and addressing interpretive ambiguities relative to the other models.The Fidelity to Legal Instrument model best aligns with modern approaches to statutory interpretation.However, it is best suited for natural-language to natural-language translation in which the functions of both the prototext and the metatext are identical.Consequently, the Fidelity to Legal Instrument approach requires an ongoing balance between legal code utility and functionality and legal alignment.Finally, the Fidelity to Function model is most receiver-oriented, allowing its methodologies and resources to be informed by the specific purpose the code is designed to meet.As a result, the Fidelity to Function model has the potential to best serve the specific needs of compliance users, but generally cannot be taken as a faithful representation of the law.
The application of fidelity models to the legal coding exercise better highlights the interpretive context of encoding legislation.The models bring with them different ways of understanding the encoding exercise and provide different methods for resolving interpretive ambiguities with reference to a range of primary and secondary materials.Further research is required to determine the contexts in which each of these approaches may be best suited, as well as how their application affects the outputs, reliability, and user experience provided by the code.Who is the code client or user, what the code is used for and how it is intended to be employed at scale are issues that are likely to significantly inform which fidelity model is best suited in a given legal coding development environment.These issues need further exploration, but it can already be said that the introduction of fidelity models based on legal translation methodologies provides much needed insight in the complexities of legal coding that require more nuanced and sophisticated interpretive approaches.

Table 1 .
Fidelity model application to legal coding exercises.