Stephen Pulman

Automatic extraction of protein interactions from scientific abstracts.

This paper motivates the use of Information Extraction (IE) for gathering data on protein interactions, describes the customization of an existing IE system, SRIs Highlight, for this task and presents the results of an experiment on unseen Medline abstracts which show that customization to a new domain can be fast, reliable and cost-effective.

A Classifier-Based Approach to Preposition and Determiner Error Correction in L2 English

In this paper, we present an approach to the automatic identification and correction of preposition and determiner errors in non-native (L2) English writing. We show that models of use for these parts of speech can be learned with an accuracy of 70.06% and 92.15% respectively on L1 text, and present first results in an error detection task for L2 writing.

Characterizing Humour: An Exploration of Features in Humorous Texts

This paper investigates the problem of automatic humour recognition, and provides and in-depth analysis of two of the most frequently observed features of humorous text: human-centeredness and negative polarity. Through experiments performed on two collections of humorous texts, we show that these properties of verbal humour are consistent across different data sets.

Concrete sentence spaces for compositional distributional models of meaning

Coecke, Sadrzadeh, and Clark [3] developed a compositional model of meaning for distributional semantics, in which each word in a sentence has a meaning vector and the distributional meaning of the sentence is a function of the tensor products of the word vectors. Abstractly speaking, this function is the morphism corresponding to the grammatical structure of the sentence in the category of finite dimensional vector spaces. In this paper, we provide a concrete method for implementing this linear meaning map, by constructing a corpus-based vector space for the type of sentence. Our construction method is based on structured vector spaces whereby meaning vectors of all sentences, regardless of their grammatical structure, live in the same vector space. Our proposed sentence space is the tensor product of two noun spaces, in which the basis vectors are pairs of words each augmented with a grammatical role. This enables us to compare meanings of sentences by simply taking the inner product of their vectors.

Automatically Acquiring Models of Preposition Use

This paper proposes a machine-learning based approach to predict accurately, given a syntactic and semantic context, which preposition is most likely to occur in that context. Each occurrence of a preposition in an English corpus has its context represented by a vector containing 307 features. The vectors are processed by a voted perceptron algorithm to learn associations between contexts and prepositions. In preliminary tests, we can associate contexts and prepositions with a success rate of up to 84.5%.

Grammars, parsers, and memory limitations

Abstract Linguistic competence cannot be adequately characterized by grammatical devices of finite state power. Nevertheless, there are reasons to suspect that the human parsing device cannot adequately deal with languages that fall outside this class. This paper discusses these issues, arguing that restrictions on available parsing memory, and on our ability to operate properly when parsing recursive constructions, mean that there is an interesting sense in which human parsing resources must be characterized as finite state. This means that certain constructions regarded as grammatical according to a (richer than finite state) competence grammar, are not parsable, or are parsed in a way which is not in exact correspondence to their description by this grammar. This raises the further question of how these constructions can nevertheless be understood appropriately, given the assumption that semantic interpretation relies on syntactic structure. The paper goes on to describe an implemented computer program...

Higher Order Unification and the Interpretation of Focus

Higher order unification is a way of combining information (or equivalently, solving equations) expressed as terms of a typed higher order logic. A suitably restricted form of the notion has been used as a simple and perspicuous basis for the resolution of the meaning of elliptical expressions and for the interpretation of some non-compositional types of comparative construction also involving ellipsis. This paper explores another area of application for this concept in the interpretation of sentences containing intonationally marked ‘focus’, or various semantic constructs which are sensitive to focus.Similarities and differences between this approach, and theories using ‘alternative semantics,’ ‘structured meanings’, or flexible categorial grammars, are described. The paper argues that the higher order unification approach offers descriptive advantages over these alternatives, as well as the practical advantage of being capable of fairly direct computational implementation.

A Method for Controlling the Production of Specifications in Natural Language

Before a system can be formally defined, it is common to write a specification in a natural language as the basis for the formal definition. Natural languages are not well suited for this task; documentation written in a natural language is often ambiguous and imprecise. Inspection of real documentation also reveals that, without special training, most writers do not produce concise, clear or consistent statements. We present here an experimental interface to a general-purpose natural-language processing system designed to control the writing of specification statements in a natural language. The interface is designed to reduce the degree of imprecision and ambiguity in the natural-language statements, as well as contributing to the writing of shorter statements, with clear structure and punctuation. We compare our approach to similar work in this area. 1. INTRODUCTIO N

Tweets about hospital quality: a mixed methods study

Background Twitter is increasingly being used by patients to comment on their experience of healthcare. This may provide information for understanding the quality of healthcare providers and improving services. Objective To examine whether tweets sent to hospitals in the English National Health Service contain information about quality of care. To compare sentiment on Twitter about hospitals with established survey measures of patient experience and standardised mortality rates. Design A mixed methods study including a quantitative analysis of all 198 499 tweets sent to English hospitals over a year and a qualitative directed content analysis of 1000 random tweets. Twitter sentiment and conventional quality metrics were compared using Spearmans rank correlation coefficient. Key results 11% of tweets to hospitals contained information about care quality, with the most frequent topic being patient experience (8%). Comments on effectiveness or safety of care were present, but less common (3%). 77% of tweets about care quality were positive in tone. Other topics mentioned in tweets included messages of support to patients, fundraising activity, self-promotion and dissemination of health information. No associations were observed between Twitter sentiment and conventional quality metrics. Conclusions Only a small proportion of tweets directed at hospitals discuss quality of care and there was no clear relationship between Twitter sentiment and other measures of quality, potentially limiting Twitter as a medium for quality monitoring. However, tweets did contain information useful to target quality improvement activity. Recent enthusiasm by policy makers to use social media as a quality monitoring and improvement tool needs to be carefully considered and subjected to formal evaluation.

Unsupervised Classification of Dialogue Acts using a Dirichlet Process Mixture Model

In recent years Dialogue Acts have become a popular means of modelling the communicative intentions of human and machine utterances in many modern dialogue systems. Many of these systems rely heavily on the availability of dialogue corpora that have been annotated with Dialogue Act labels. The manual annotation of dialogue corpora is both tedious and expensive. Consequently, there is a growing interest in unsupervised systems that are capable of automating the annotation process. This paper investigates the use of a Dirichlet Process Mixture Model as a means of clustering dialogue utterances in an unsupervised manner. These clusters can then be analysed in terms of the possible Dialogue Acts that they might represent. The results presented here are from the application of the Dirichlet Process Mixture Model to the Dihana corpus.