Allan Ramsay

Constructing a Normal Form for Property Theory

Property Theory, introduced in [Turner 87], is a classical first-order logic that includes a ⋅ operator to turn propositions, properties and relations into terms. It is therefore an appropriate representation language for intensional concepts such as knowledge and belief. The main advantage of Property Theory over languages like Montague semantics is that it is a type-free language, and hence provides considerable extra expressive power. The pay-back is that it is consequently extremely intractable, and constructing an appropriate normal form has proven to be very difficult.

Generating relevant models

Manthey and Brys model generation approach to theorem proving for FOPC has been greeted with considerable interest. Unfortunately the original presentation of the technique can become arbitrarily inefficient when applied to problems whose statements contain large amounts of irrelevant information. We show how to avoid these problems whilst retaining nearly all the advantages of the basic approach.

Theorem proving for untyped constructive λ-calculus: implementation and application

This paper presents a theorem prover for a highly intensional logic, namely a constructive version of property theory [25] (this language essentially provides a combination of constructive rst-order logic and the -calculus). The paper presents the basic theorem prover, which is a higher-order extension of Manthey and Bry’s model generation theorem prover for rst-order logic [14]; considers issues relating to the compile-time optimisations that are often used with rst-order theorem provers; and shows how the resulting system can be used in a natural language understanding system.

Towards including prosody in a text-to-speech system for modern standard Arabic

Most attempts to provide text-to-speech for modern standard Arabic (MSA) have concentrated on solving the problem of diacritic assignment (i.e. of recovering phonetically relevant information, such as choice of short vowels, which is not explicitly provided in the surface form of MSA). This is clearly a crucial issue: you can hardly produce intelligible spoken output if you do not know what the vowels are. We describe an approach to the task of generating speech from MSA text which not only solves this initial problem, but also provides the information required for imposing an appropriate intonation contour.

Sarcasm, Deception, and Stating the Obvious: Planning Dialogue without Speech Acts

This paper presents an alternative to the ‘speech acts with STRIPS’ approach to implementing dialogue a fully implemented AI planner which generates and analyses the semantics of utterances using a single linguistic act for all contexts. Using this act, the planner can model problematic conversational situations, including felicitous and infelicitous instances of bluffing, lying, sarcasm, and stating the obvious. The act has negligible effects, and its precondition can always be proved. ‘Speaker maxims’ enable the speaker to plan to deceive, as well as to generate implicatures, while ‘hearer maxims’ enable the hearer to recognise deceptions, and interpret implicatures. The planner proceeds by achieving parts of the constructive proof of a goal. It incorporates an epistemic theorem prover, which embodies a deduction model of belief, and a constructive logic.

Parsing with discontinuous phrases

Most parsing algorithms require phrases that are to be combined to be either contiguous or marked as being ‘extraposed’. The assumption that phrases which are to be combined will be adjacent to one another supports rapid indexing mechanisms: the fact that in most languages items can turn up in unexpected locations cancels out much of the ensuing efficiency. The current paper shows how ‘out of position’ items can be incorporated directly. This leads to efficient parsing even when items turn up having been right-shifted, a state of affairs which makes Johnson and Kays (1994) notion of ‘sponsorship’ of empty nodes inapplicable.

Natural language inference for arabic using extended tree edit distance with subtrees

Many natural language processing (NLP) applications require the computation of similarities between pairs of syntactic or semantic trees. Many researchers have used tree edit distance for this task, but this technique suffers from the drawback that it deals with single node operations only. We have extended the standard tree edit distance algorithm to deal with subtree transformation operations as well as single nodes. The extended algorithm with subtree operations, TED+ST, is more effective and flexible than the standard algorithm, especially for applications that pay attention to relations among nodes (e.g. in linguistic trees, deleting a modifier subtree should be cheaper than the sum of deleting its components individually). We describe the use of TED+ST for checking entailment between two Arabic text snippets. The preliminary results of using TED+ST were encouraging when compared with two string-based approaches and with the standard algorithm.

Local constraints on arabic word order

Syntactic analysis of Arabic poses two major problems. (i) Although the canonical order of Arabic sentences is VSO, a range of other orders are possible. In order to carry out such an analysis, then, it is necessary to have a grammatical framework and an associated parsing algorithm that can cope with free word order. (ii) Although a range of non-canonical orders are possible, not all orders are possible under all circumstances. The current paper outlines an approach to obtaining syntactic descriptions of sentences of Modern Standard Arabic, where the problems outlined above are compounded by omission of short vowels and other critical information from the written form.

Speech Acts, Epistemic Planning and Grice's Maxims

Work on speech acts has generally involved the introduction of sets of different actions such as informing, reminding, bluffing and lying. These actions have different preconditions and effects, and hence can be used to achieve a wide variety of different real-world goals. The problem is that they tend to have indistinguishable surface forms. As such, it is extremely difficult for the hearer to decide which action she thinks has been performed, and it is therefore also extremely difficult for the speaker to be confident about how the hearer will respond. We will show how to achieve complex goals on the basis of a very simple set of linguistic actions. These actions have clearly marked surface forms, and hence can easily be distinguishable by a hearer. In order to do this, we have developed an epistemic planner with a number of interesting features, and with a number of optimisations that relate directly to aspects of the task at hand.

WEAK LEXICAL SEMANTICS AND MULTIPLE VIEWS

Most formal/computational treatments of natural language semantics discuss the production of formal paraphrases in some suitable language. There is some debate as to whether this can be done strictly compositionally, e.g. by function application and composition, as proposed in Montague grammar and its descendants (Dowty et al., 1981; Kohlhase et al., 1996), or by simple structure sharing (Pollard and Sag, 1988; Pollard and Sag, 1994; Fenstad et al., 1987); or whether some more indirect construction process is involved (Dalrymple et al., 1996; van Genabith and Crouch, 1997). In nearly every case, however, the goal of the process is a formal paraphrase that ‘means the same as the original’.