We present an in-depth comparison of three clinical information extraction (IE) systems designed to perform entity recognition and negation detection on brain imaging reports: EdIE-R, a bespoke rule-based system, and two neural network models, EdIE-BiLSTM and EdIE-BERT, both multi-task learning models with a BiLSTM and BERT encoder respectively. We compare our models both on an in-sample and an out-of-sample dataset containing mentions of stroke findings and draw on our error analysis to suggest improvements for effective annotation when building clinical NLP models for a new domain. Our analysis finds that our rule-based system outperforms the neural models on both datasets and seems to generalise to the out-of-sample dataset. On the other hand, the neural models do not generalise negation to the out-of-sample dataset, despite metrics on the in-sample dataset suggesting otherwise.

This paper describes work in progress on devising automatic and parallel methods for geoparsing large digital historical textual data by combining the strengths of three natural language processing (NLP) tools, the Edinburgh Geoparser, spaCy and defoe, and employing different tokenisation and named entity recognition (NER) techniques. We apply these tools to a large collection of nineteenth century Scottish geographical dictionaries, and describe preliminary results obtained when processing this data.

Twitter-related studies often need to geo-locate Tweets or Twitter users, identifying their real-world geographic locations. As tweet-level geotagging remains rare, most prior work exploited tweet content, timezone and network information to inform geolocation, or else relied on off-the-shelf tools to geolocate users from location information in their user profiles. However, such user location metadata is not consistently structured, causing such tools to fail regularly, especially if a string contains multiple locations, or if locations are very fine-grained. We argue that user profile location (UPL) and tweet location need to be treated as distinct types of information from which differing inferences can be drawn. Here, we apply geoparsing to UPLs, and demonstrate how task performance can be improved by adapting our Edinburgh Geoparser, which was originally developed for processing English text. We present a detailed evaluation method and results, including inter-coder agreement. We demonstrate that the optimised geoparser can effectively extract and geo-reference multiple locations at different levels of granularity with an F1-score of around 0.90. We also illustrate how geoparsed UPLs can be exploited for international information trade studies and country-level sentiment analysis.

This work investigates how automated methods can be used to classify social media text into argumentation types. In particular it is shown how supervised machine learning was used to annotate a Twitter dataset (London Riots) with argumentation classes. An investigation of issues arising from a natural inconsistency within social media data found that machine learning algorithms tend to over fit to the data because Twitter contains a lot of repetition in the form of retweets. It is also noted that when learning argumentation classes we must be aware that the classes will most likely be of very different sizes and this must be kept in mind when analysing the results. Encouraging results were found in adapting a model from one domain of Twitter data (London Riots) to another (OR2012). When adapting a model to another dataset the most useful feature was punctuation. It is probable that the nature of punctuation in Twitter language, the very specific use in links, indicates argumentation class.

In biomedical articles, terms with the same surface forms are often used to refer to different entities across a number of model organisms, in which case determining the species becomes crucial to term identification systems that ground terms to specific database identifiers. This paper describes a rule-based system that extracts “species indicating words”, such as human or murine, which can be used to decide the species of the nearby entity terms, and a machine-learning species disambiguation system that was developed on manually species-annotated corpora. Performance of both systems were evaluated on gold-standard datasets, where the machine-learning system yielded better overall results.

We describe and evaluate a prototype system for recognising person and place names in digitised records of British parliamentary proceedings from the late 17th and early 19th centuries. The output of an OCR engine is the input for our system and we describe certain issues and errors in this data and discuss the methods we have used to overcome the problems. We describe our rule-based named entity recognition system for person and place names which is implemented using the LT-XML2 and LT-TTT2 text processing tools. We discuss the annotation of a development and testing corpus and provide results of an evaluation of our system on the test corpus.

In this paper we discuss five different corpora annotated forprotein names. We present several within- and cross-dataset proteintagging experiments showing that different annotation schemes severelyaffect the portability of statistical protein taggers. By means of adetailed error analysis we identify crucial annotation issues thatfuture annotation projects should take into careful consideration.

In this paper we discuss a rule-based approach to chunking implemented using the LT-XML2 and LT-TTT2 tools. We describe the tools and the pipeline and grammars that have been developed for the task of chunking. We show that our rule-based approach is easy to adapt to different chunking styles and that the mark-up of further linguistic information such as nominal and verbal heads can be added to the rules at little extra cost. We evaluate our chunker against the CoNLL 2000 data and discuss discrepancies between our output and the CoNLL mark-up as well as discrepancies within the CoNLL data itself. We contrast our results with the higher scores obtained using machine learning and argue that the portability and flexibility of our approach still make it a more practical solution.