Controlled Natural Language

We have developed a methodology for representation of ontologies in a strictly typed language with dependent types. The methodology is supported by an experiment where we translated SUMO (Suggested Upper-Merged Ontology) to GF (Grammatical Framework). The representation of SUMO in GF preserves the expressivity of the original ontology, adding to this the advantages of a type system and built-in support for natural language generation. SUMO is the largest open-source ontology describing over 10,000 concepts and the relations between them, along with a number of first-order axioms, which are further on used in performing automated reasoning on the ontology. GF is a type-theoretical grammar formalism mainly used for natural language applications. Through the logical framework that it incorporates, GF allows a consistent ontology representation, and thanks to its grammatical features the ontology is directly verbalized in a number of controlled natural languages.

Legislative language exhibits some characteristics typical of languages of administration that are particularly prone to eliciting ambiguities. However, ambiguity is generally undesirable in legislative texts and can pose problems for the interpretation and application of codified law. In this paper, we demonstrate how methods of controlled natural languages can be applied to prevent ambiguities in legislative texts. We investigate what types of ambiguities are frequent in legislative language and therefore important to control, and we examine which ambiguities are already controlled by existing drafting guidelines. For those not covered by the guidelines, we propose additional control mechanisms. Wherever possible, the devised mechanisms reflect existing conventions and frequency distributions and exploit domain-specific means to make ambiguities explicit.

The Naproche CNL is a controlled natural language for mathematical texts. A recent addition to the Naproche CNL are plural statements. We discuss the collective-distributive ambiguity in the context of mathematical language, as well as pairwise interpretations of collective plurals. Additionally, we present a special scope ambiguity conjunctions give rise to. Finally, we describe an innovative plural interpretation algorithm implemented in Naproche for disambiguating plurals in DRT and giving them the interpretation that would normally be preferred in a mathematical context.

The Semantic Web is yet to gain mainstream recognition. In part this is caused by the relative complexity of the various semantic web formalisms, which act as a major barrier of entry to naive web users. In addition, in order for the Semantic Web to become a reality, we need semantic metadata. While controlled natural language research has sought to address these challenges, in the context of user friendly ontology authoring for domain experts, there has been little focus on how to adapt controlled languages for novice social web users. The paper describes an approach to using controlled languages for fact creation and management as opposed to ontology authoring, focusing on the domain of meeting minutes. For demonstration purposes, we developed a plug-in to the Semantic MediaWiki, which adds a controlled language editor extension. This editor aids the user while authoring or annotating in a controlled language in a user friendly manner. Controlled content is sent to a parsing service which generates semantic metadata from the sentences which are subsequently displayed and stored in the Semantic MediaWiki. The semantic metadata generated by the parser is grounded against a project documents ontology. The controlled language modeled covers a wide variety of sentences and topics used in the context of a meeting minute. Finally this paper provides a architectural overview of the annotation system.

RACE is a first-order reasoner for Attempto Controlled English (ACE) that can show the (in-) consistency of a set of ACE axioms, prove ACE theorems from ACE axioms and answer ACE queries from ACE axioms. In each case RACE gives a proof justification in ACE and full English. This paper is a system description of RACE sketching its structure, its implementation, its operation and its user interface. The power and the limitations of RACE are demonstrated and discussed by concrete examples.

This paper introduces a new grammar notation, called Codeco, designed for controlled natural language (CNL) and predictive editors. Existing grammar frameworks that target either formal or natural languages do not work out particularly well for CNL, especially if they are to be used in predictive editors and if anaphoric references should be resolved in a deterministic way. It is not trivial to build predictive editors that can precisely determine which anaphoric references are possible at a certain position. This paper shows how such complex structures can be represented in Codeco, a novel grammar notation for CNL. Two different parsers have been implemented (one in Prolog and another one in Java) and a large subset of Attempto Controlled English (ACE) has been represented in Codeco. The results show that Codeco is practical, adequate and efficient.

Controlled languages are usually targeted for technical domains and designed to be unambiguous. This paper presents a controlled language whose domain is touristic phrases, aimed to be usable by anyone without prior training. Despite its informal nature, the language of phrases has a firm notion of semantics, defining the correctness of translations. However, this semantics is formulated in terms of context and situation rather than by logical formulas. Moreover, the language is often ambiguous, and the translation may depend on resolving the ambiguity. This paper shows how to formalize a semantics for tourist phrases and implement it in 15 languages, how to deal with the ambiguities, and how to make the system available for layman users on the web and on mobile phones. While a useful application as such, the Phrasebook also paves the way for an extended notion of controlled language, and the techniques are aimed to be general enough to support many such extensions.

This paper addresses the design of an automated legal assistant capable of performing a logical analysis of legal documents and using natural language as a medium of communication with a human client. We focus on the interplay between natural language in which the legal document is expressed and the formal logic used for reasoning about it — ideally approached using a controlled natural language (CNL) together with an appropriately chosen logic for analysis and reasoning. In translating from CNL to logic, information about the CNL structure is lost. For example, the CNL might contain legal clause numbers, whilst the logic might not. This can lead to problems when for example the reasoning system discovers an inconsistency in the contract and needs to explain its whereabouts to the client. Below we discuss the issues affecting the choice of logic, arguing in favour of keeping certain structural information during formal analysis of legal documents to be able to refer to that structure when interacting with the user.

We present a framework in which to experiment and seek solutions to these issues. Having identified a sufficiently restricted domain of application we also report on the development of a CNL to interact with a variant of the game Nomic — a game based on the notion of contract specification and amendment — and argue how this game provides an ideal platform to explore the use of structure information in the domain of legal analysis.

In this paper I discuss how the controlled natural language PENG Light can be modified so that it can serve as a high-level interface language to the Event Calculus. The Event Calculus is a narrative-based formal language for reasoning about events, their effects and timepoints, and can be used for various reasoning tasks where a representation of time is important. Using a scenario from a dynamic domain, I show what kind of modifications are necessary on the level of the controlled natural language to specify the background knowledge that is required to deal with direct and indirect effects of events and with continuous change in that domain. I discuss how the output of the controlled natural language processor of PENG Light that distinguishes between events and states can be aligned with the input language of the Event Calculus, and then be used for automated reasoning. Finally, I show how the Event Calculus can be used to support the question answering process and then evaluate its reasoning capabilities using a number of benchmark questions stated in controlled natural language.

Involving domain experts in modeling is important when knowledge needs to be captured in a model and only domain experts can establish whether the models are correct. We have experienced that a natural language based representation of a model helps them to understand the semantics of a model and has advantages over a visual representation. Therefore a controlled natural language (CNL) is designed for our existing semantic reasoning tool Be Informed, which is based on conceptual graphs. The resulting CNL has a formal logical basis but the goal of the CNL representation is to improve readability for human readers. We report on the challenge to develop a CNL that 1) is easy and intuitively readable for domain experts with no background in formal logics, 2) can be easily generated from the formal representation and 3) can be easily adjusted for other natural languages and cultural preferences. The solution uses patterns to represent the CNL that map to the conceptual graph. The patterns are based on SBVR’s RuleSpeak and can be easily adjusted for local differences.