We introduce a new dataset for machine translation of Guinea-Bissau Creole (Kiriol), comprising around 40 thousand parallel sentences to English and Portuguese. This dataset is made up of predominantly religious data (from the Bible and texts from the Jehovah’s Witnesses), but also a small amount of general domain data (from a dictionary). This mirrors the typical resource availability of many low resource languages. We train a number of transformer-based models to investigate how to improve domain transfer from religious data to a more general domain. We find that adding even 300 sentences from the target domain when training substantially improves the translation performance, highlighting the importance and need for data collection for low-resource languages, even on a small-scale. We additionally find that Portuguese-to-Kiriol translation models perform better on average than other source and target language pairs, and investigate how this relates to the morphological complexity of the languages involved and the degree of lexical overlap between creoles and lexifiers. Overall, we hope our work will stimulate research into Kiriol and into how machine translation might better support creole languages in general.

Igbo is a low-resource language spoken by approximately 30 million people worldwide. It is the native language of the Igbo people of south-eastern Nigeria. In Igbo language, diacritics - orthographic and tonal - play a huge role in the distinguishing the meaning and pronunciation of words. Omitting diacritics in texts often leads to lexical ambiguity. Diacritic restoration is a pre-processing task that replaces missing diacritics on words from which they have been removed. In this work, we applied embedding models to the diacritic restoration task and compared their performances to those of n-gram models. Although word embedding models have been successfully applied to various NLP tasks, it has not been used, to our knowledge, for diacritic restoration. Two classes of word embeddings models were used: those projected from the English embedding space; and those trained with Igbo bible corpus (≈ 1m). Our best result, 82.49%, is an improvement on the baseline n-gram models.

Existing NLP models are mostly trained with data from well-resourced languages. Most minority languages face the challenge of lack of resources - data and technologies - for NLP research. Building these resources from scratch for each minority language will be very expensive, time-consuming and amount largely to unnecessarily re-inventing the wheel. In this paper, we applied transfer learning techniques to create Igbo word embeddings from a variety of existing English trained embeddings. Transfer learning methods were also used to build standard datasets for Igbo word similarity and analogy tasks for intrinsic evaluation of embeddings. These projected embeddings were also applied to diacritic restoration task. Our results indicate that the projected models not only outperform the trained ones on the semantic-based tasks of analogy, word-similarity, and odd-word identifying, but they also achieve enhanced performance on the diacritic restoration with learned diacritic embeddings.

Properly written texts in Igbo, a low-resource African language, are rich in both orthographic and tonal diacritics. Diacritics are essential in capturing the distinctions in pronunciation and meaning of words, as well as in lexical disambiguation. Unfortunately, most electronic texts in diacritic languages are written without diacritics. This makes diacritic restoration a necessary step in corpus building and language processing tasks for languages with diacritics. In our previous work, we built some n-gram models with simple smoothing techniques based on a closed-world assumption. However, as a classification task, diacritic restoration is well suited for and will be more generalisable with machine learning. This paper, therefore, presents a more standard approach to dealing with the task which involves the application of machine learning algorithms.

Automatic summarization of reader comments in on-line news is an extremely challenging task and a capability for which there is a clear need. Work to date has focussed on producing extractive summaries using well-known techniques imported from other areas of language processing. But are extractive summaries of comments what users really want? Do they support users in performing the sorts of tasks they are likely to want to perform with reader comments? In this paper we address these questions by doing three things. First, we offer a specification of one possible summary type for reader comment, based on an analysis of reader comment in terms of issues and viewpoints. Second, we define a task-based evaluation framework for reader comment summarization that allows summarization systems to be assessed in terms of how well they support users in a time-limited task of identifying issues and characterising opinion on issues in comments. Third, we describe a pilot evaluation in which we used the task-based evaluation framework to evaluate a prototype reader comment clustering and summarization system, demonstrating the viability of the evaluation framework and illustrating the sorts of insight such an evaluation affords.

Streaming media provides a number of unique challenges for computational linguistics. This paper studies the temporal variation in word co-occurrence statistics, with application to event detection. We develop a spectral clustering approach to find groups of mutually informative terms occurring in discrete time frames. Experiments on large datasets of tweets show that these groups identify key real world events as they occur in time, despite no explicit supervision. The performance of our method rivals state-of-the-art methods for event detection on F-score, obtaining higher recall at the expense of precision.

Lexical substitution is the task of finding a replacement for a target word in a sentence so as to preserve, as closely as possible, the meaning of the original sentence. It has been proposed that lexical substitution be used as a basis for assessing the performance of word sense disambiguation systems, an idea realised in the English Lexical Substitution Task of SemEval-2007. In this paper, we examine the evaluation metrics used for the English Lexical Substitution Task and identify some problems that arise for them. We go on to propose some alternative measures for this purpose, that avoid these problems, and which in turn can be seen as redefining the key tasks that lexical substitution systems should be expected to perform. We hope that these new metrics will better serve to guide the development of lexical substitution systems in future work. One of the new metrics addresses how effective systems are in ranking substitution candidates, a key ability for lexical substitution systems, and we report some results concerning the assessment of systems produced by this measure as compared to the relevant measure from SemEval-2007.

The availability of large collections of text have made it possible to build language models that incorporate counts of billions of n-grams. This paper proposes two new methods of efficiently storing large language models that allow O(1) random access and use significantly less space than all known approaches. We introduce two novel data structures that take advantage of the distribution of n-grams in corpora and make use of various numbers of minimal perfect hashes to compactly store language models containing full frequency counts of billions of n-grams using 2.5 Bytes per n-gram and language models of quantized probabilities using 2.26 Bytes per n-gram. These methods allow language processing applications to take advantage of much larger language models than previously was possible using the same hardware and we additionally describe how they can be used in a distributed environment to store even larger models. We show that our approaches are simple to implement and can easily be combined with pruning and quantization to achieve additional reductions in the size of the language model.

Terminologies and other knowledge resources are widely used to aid entity recognition in specialist domain texts. As well as providing lexicons of specialist terms, linkage from the text back to a resource can make additional knowledge available to applications. Use of such resources is especially pertinent in the biomedical domain, where large numbers of these resources are available, and where they are widely used in informatics applications. Terminology resources can be most readily used by simple lexical lookup of terms in the text. A major drawback with such lexical lookup, however, is poor precision caused by ambiguity between domain terms and general language words. We combine lexical lookup with simple filtering of ambiguous terms, to improve precision. We compare this lexical lookup with a statistical method of entity recognition, and to a method which combines the two approaches. We show that the combined method boosts precision with little loss of recall, and that linkage from recognised entities back to the domain knowledge resources can be maintained.

We present recent work in the area of Cross-Domain Dialogue Act (DA) tagging. We have previously reported on the use of a simple dialogue act classifier based on purely intra-utterance features - principally involving word n-gram cue phrases automatically generated from a training corpus. Such a classifier performs surprisingly well, rivalling scores obtained using far more sophisticated language modelling techniques. In this paper, we apply these automatically extracted cues to a new annotated corpus, to determine the portability and generality of the cues we learn.