FIN-CLARIN: A Framework for Depositing and Disseminating Language Resources for R&D

Researchers and developers in academia and industry would benefit from a facility that enables them to easily locate, licence and use the kind of empirical data they need for testing and refining their hypotheses and to deposit and disseminate their data e.g. to support replication and validation of reported scientific experiments. To answer these needs initially in Finland, there is an ongoing project at University of Helsinki and its collaborators to create a user-friendly web service for researchers and developers in Finland and other countries. In our talk, we describe ongoing work to create a palette of extensive but easily available Finnish language resources and technologies for the research community, including lexical resources, wordnets, morphologically tagged corpora, dependency syntactic treebanks and parsebanks, open-source finite state toolkits and libraries and language models to support text analysis and processing at customer site. Also first publicly available results are presented.

Discriminative language model adaptation for Mandarin broadcast speech transcription and translation

This paper investigates unsupervised test-time adaptation of language models (LM) using discriminative methods for a Mandarin broadcast speech transcription and translation task. A standard approach to adapt interpolated language models to is to optimize the component weights by minimizing the perplexity on supervision data. This is a widely made approximation for language modeling in automatic speech recognition (ASR) systems. For speech translation tasks, it is unclear whether a strong correlation still exists between perplexity and various forms of error cost functions in recognition and translation stages. The proposed minimum Bayes risk (MBR) based approach provides a flexible framework for unsupervised LM adaptation. It generalizes to a variety of forms of recognition and translation error metrics. LM adaptation is performed at the audio document level using either the character error rate (CER), or translation edit rate (TER) as the cost function. An efficient parameter estimation scheme using the extended Baum-Welch (EBW) algorithm is proposed. Experimental results on a state-of-the-art speech recognition and translation system are presented. The MBR adapted language models gave the best recognition and translation performance and reduced the TER score by up to 0.54% absolute. © 2007 IEEE.

Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple subsystems developed at different sites. Cross system adaptation can be used as an alternative to direct hypothesis level combination schemes such as ROVER. In normal cross adaptation it is assumed that useful diversity among systems exists only at acoustic level. However, complimentary features among complex LVCSR systems also manifest themselves in other layers of modelling hierarchy, e.g., subword and word level. It is thus interesting to also cross adapt language models (LM) to capture them. In this paper cross adaptation of multi-level LMs modelling both syllable and word sequences was investigated to improve LVCSR system combination. Significant error rate gains up to 6.7% rel. were obtained over ROVER and acoustic model only cross adaptation when combining 13 Chinese LVCSR subsystems used in the 2010 DARPA GALE evaluation. © 2010 ISCA.

Use of contexts in language model interpolation and adaptation

Language models (LMs) are often constructed by building multiple individual component models that are combined using context independent interpolation weights. By tuning these weights, using either perplexity or discriminative approaches, it is possible to adapt LMs to a particular task. This paper investigates the use of context dependent weighting in both interpolation and test-time adaptation of language models. Depending on the previous word contexts, a discrete history weighting function is used to adjust the contribution from each component model. As this dramatically increases the number of parameters to estimate, robust weight estimation schemes are required. Several approaches are described in this paper. The first approach is based on MAP estimation where interpolation weights of lower order contexts are used as smoothing priors. The second approach uses training data to ensure robust estimation of LM interpolation weights. This can also serve as a smoothing prior for MAP adaptation. A normalized perplexity metric is proposed to handle the bias of the standard perplexity criterion to corpus size. A range of schemes to combine weight information obtained from training data and test data hypotheses are also proposed to improve robustness during context dependent LM adaptation. In addition, a minimum Bayes' risk (MBR) based discriminative training scheme is also proposed. An efficient weighted finite state transducer (WFST) decoding algorithm for context dependent interpolation is also presented. The proposed technique was evaluated using a state-of-the-art Mandarin Chinese broadcast speech transcription task. Character error rate (CER) reductions up to 7.3 relative were obtained as well as consistent perplexity improvements. © 2012 Elsevier Ltd. All rights reserved.

Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple sub-systems that may even be developed at different sites. Cross system adaptation, in which model adaptation is performed using the outputs from another sub-system, can be used as an alternative to hypothesis level combination schemes such as ROVER. Normally cross adaptation is only performed on the acoustic models. However, there are many other levels in LVCSR systems' modelling hierarchy where complimentary features may be exploited, for example, the sub-word and the word level, to further improve cross adaptation based system combination. It is thus interesting to also cross adapt language models (LMs) to capture these additional useful features. In this paper cross adaptation is applied to three forms of language models, a multi-level LM that models both syllable and word sequences, a word level neural network LM, and the linear combination of the two. Significant error rate reductions of 4.0-7.1% relative were obtained over ROVER and acoustic model only cross adaptation when combining a range of Chinese LVCSR sub-systems used in the 2010 and 2011 DARPA GALE evaluations. © 2012 Elsevier Ltd. All rights reserved.

Exploratory study : exploring the effectiveness of a literacy-based intervention with children with language impairments

Research points clearly to the need for all concerned stakeholders to adopt a preventative approach while intervening with children who are at-risk for future reading disabilities. Research has indicated also that a particular sub-group of children at-risk for reading impairments include preschool children with language impairments (Catts, 1993). Preschool children with language impairments may have difficulties with emergent literacy skills - important prerequisite skills necessary for successful formal reading. Only in the past decade have researchers begun to study the effects of emergent literacy intervention on preschool children with language impairments. As such, the current study continues this investigation of how to effectively implement an emergent literacy therapy aimed at supporting preschool children with language impairments. In addition to this, the current study explores emergent literacy intervention within an applied clinical setting. The setting, presents a host of methodological and theoretical challenges - challenges that will advance the field of understanding children within naturalistic settings. This exploratory study included thirty-eight participants who were recruited from Speech Services Niagara, a local preschool speech and language program. Using a between-group pre- and posttest design, this study compared two intervention approaches - an experimental emergent literacy intervention and a traditional language intervention. The experimental intervention was adopted from Read It Again! (Justice, McGinty, Beckman, & Kilday, 2006) and the traditional language intervention was based on the traditional models of language therapy typically used in preschool speech and language models across Ontario. 5 Results indicated that the emergent literacy intervention was superior to the ,t..3>~, ~\., ;./h traditional language therapy in improving the children's alphabet knowledge, print and word awareness and phonological awareness. Moreover, results revealed that children with more severe language impairments require greater support and more explicit instruction than children with moderate language impairments. Another important finding indicated that the effects of the preschool emergent literacy intervention used in this study may not be sustainable as children enter grade one. The implications of this study point to the need to support preschool children with language impairments with intensive emergent literacy intervention that extends beyond preschool into formal educational settings.

Análise de proteção de linhas de transmissão através de relés numéricos e uso de models externos no ATP (Alternative Transient Program)

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

Language identification based on a discriminative text categorization technique

In this paper, we describe new results and improvements to a lan-guage identification (LID) system based on PPRLM previously introduced in [1] and [2]. In this case, we use as parallel phone recognizers the ones provided by the Brno University of Technology for Czech, Hungarian, and Russian lan-guages, and instead of using traditional n-gram language models we use a lan-guage model that is created using a ranking with the most frequent and discrim-inative n-grams. In this language model approach, the distance between the ranking for the input sentence and the ranking for each language is computed, based on the difference in relative positions for each n-gram. This approach is able to model reliably longer span information than in traditional language models obtaining more reliable estimations. We also describe the modifications that we have being introducing along the time to the original ranking technique, e.g., different discriminative formulas to establish the ranking, variations of the template size, the suppression of repeated consecutive phones, and a new clus-tering technique for the ranking scores. Results show that this technique pro-vides a 12.9% relative improvement over PPRLM. Finally, we also describe re-sults where the traditional PPRLM and our ranking technique are combined.

Analysis of the effect of negation on information retrieval of medical data

Most information retrieval (IR) models treat the presence of a term within a document as an indication that the document is somehow "about" that term, they do not take into account when a term might be explicitly negated. Medical data, by its nature, contains a high frequency of negated terms - e.g. "review of systems showed no chest pain or shortness of breath". This papers presents a study of the effects of negation on information retrieval. We present a number of experiments to determine whether negation has a significant negative affect on IR performance and whether language models that take negation into account might improve performance. We use a collection of real medical records as our test corpus. Our findings are that negation has some affect on system performance, but this will likely be confined to domains such as medical data where negation is prevalent.

TOPSIG : Topology Preserving Document Signatures

Performance comparisons between File Signatures and Inverted Files for text retrieval have previously shown several significant shortcomings of file signatures relative to inverted files. The inverted file approach underpins most state-of-the-art search engine algorithms, such as Language and Probabilistic models. It has been widely accepted that traditional file signatures are inferior alternatives to inverted files. This paper describes TopSig, a new approach to the construction of file signatures. Many advances in semantic hashing and dimensionality reduction have been made in recent times, but these were not so far linked to general purpose, signature file based, search engines. This paper introduces a different signature file approach that builds upon and extends these recent advances. We are able to demonstrate significant improvements in the performance of signature file based indexing and retrieval, performance that is comparable to that of state of the art inverted file based systems, including Language models and BM25. These findings suggest that file signatures offer a viable alternative to inverted files in suitable settings and positions the file signatures model in the class of Vector Space retrieval models.

Retrieval of health advice on the web: AEHRC at ShARe/CLEF eHealth Evaluation Lab Task 3

This paper details the participation of the Australian e- Health Research Centre (AEHRC) in the ShARe/CLEF 2013 eHealth Evaluation Lab { Task 3. This task aims to evaluate the use of information retrieval (IR) systems to aid consumers (e.g. patients and their relatives) in seeking health advice on the Web. Our submissions to the ShARe/CLEF challenge are based on language models generated from the web corpus provided by the organisers. Our baseline system is a standard Dirichlet smoothed language model. We enhance the baseline by identifying and correcting spelling mistakes in queries, as well as expanding acronyms using AEHRC's Medtex medical text analysis platform. We then consider the readability and the authoritativeness of web pages to further enhance the quality of the document ranking. Measures of readability are integrated in the language models used for retrieval via prior probabilities. Prior probabilities are also used to encode authoritativeness information derived from a list of top-100 consumer health websites. Empirical results show that correcting spelling mistakes and expanding acronyms found in queries signi cantly improves the e ectiveness of the language model baseline. Readability priors seem to increase retrieval e ectiveness for graded relevance at early ranks (nDCG@5, but not precision), but no improvements are found at later ranks and when considering binary relevance. The authoritativeness prior does not appear to provide retrieval gains over the baseline: this is likely to be because of the small overlap between websites in the corpus and those in the top-100 consumer-health websites we acquired.

Integrating and evaluating neural word embeddings in information retrieval

Recent advances in neural language models have contributed new methods for learning distributed vector representations of words (also called word embeddings). Two such methods are the continuous bag-of-words model and the skipgram model. These methods have been shown to produce embeddings that capture higher order relationships between words that are highly effective in natural language processing tasks involving the use of word similarity and word analogy. Despite these promising results, there has been little analysis of the use of these word embeddings for retrieval. Motivated by these observations, in this paper, we set out to determine how these word embeddings can be used within a retrieval model and what the benefit might be. To this aim, we use neural word embeddings within the well known translation language model for information retrieval. This language model captures implicit semantic relations between the words in queries and those in relevant documents, thus producing more accurate estimations of document relevance. The word embeddings used to estimate neural language models produce translations that differ from previous translation language model approaches; differences that deliver improvements in retrieval effectiveness. The models are robust to choices made in building word embeddings and, even more so, our results show that embeddings do not even need to be produced from the same corpus being used for retrieval.

Unrestricted Kannada online handwritten akshara recognition using SDTW

In this paper, we present an unrestricted Kannada online handwritten character recognizer which is viable for real time applications. It handles Kannada and Indo-Arabic numerals, punctuation marks and special symbols like $, &, # etc, apart from all the aksharas of the Kannada script. The dataset used has handwriting of 69 people from four different locations, making the recognition writer independent. It was found that for the DTW classifier, using smoothed first derivatives as features, enhanced the performance to 89% as compared to preprocessed co-ordinates which gave 85%, but was too inefficient in terms of time. To overcome this, we used Statistical Dynamic Time Warping (SDTW) and achieved 46 times faster classification with comparable accuracy i.e. 88%, making it fast enough for practical applications. The accuracies reported are raw symbol recognition results from the classifier. Thus, there is good scope of improvement in actual applications. Where domain constraints such as fixed vocabulary, language models and post processing can be employed. A working demo is also available on tablet PC for recognition of Kannada words.

Performance enhancement of online handwritten Tamil symbol recognition with reevaluation techniques

In this article, we aim at reducing the error rate of the online Tamil symbol recognition system by employing multiple experts to reevaluate certain decisions of the primary support vector machine classifier. Motivated by the relatively high percentage of occurrence of base consonants in the script, a reevaluation technique has been proposed to correct any ambiguities arising in the base consonants. Secondly, a dynamic time-warping method is proposed to automatically extract the discriminative regions for each set of confused characters. Class-specific features derived from these regions aid in reducing the degree of confusion. Thirdly, statistics of specific features are proposed for resolving any confusions in vowel modifiers. The reevaluation approaches are tested on two databases (a) the isolated Tamil symbols in the IWFHR test set, and (b) the symbols segmented from a set of 10,000 Tamil words. The recognition rate of the isolated test symbols of the IWFHR database improves by 1.9 %. For the word database, the incorporation of the reevaluation step improves the symbol recognition rate by 3.5 % (from 88.4 to 91.9 %). This, in turn, boosts the word recognition rate by 11.9 % (from 65.0 to 76.9 %). The reduction in the word error rate has been achieved using a generic approach, without the incorporation of language models.

Recognition of open vocabulary, online handwritten pages in Tamil script

In this work, we describe a system, which recognises open vocabulary, isolated, online handwritten Tamil words and extend it to recognize a paragraph of writing. We explain in detail each step involved in the process: segmentation, preprocessing, feature extraction, classification and bigram-based post-processing. On our database of 45,000 handwritten words obtained through tablet PC, we have obtained symbol level accuracy of 78.5% and 85.3% without and with the usage of post-processing using symbol level language models, respectively. Word level accuracies for the same are 40.1% and 59.6%. A line and word level segmentation strategy is proposed, which gives promising results of 100% line segmentation and 98.1% word segmentation accuracies on our initial trials of 40 handwritten paragraphs. The two modules have been combined to obtain a full-fledged page recognition system for online handwritten Tamil data. To the knowledge of the authors, this is the first ever attempt on recognition of open vocabulary, online handwritten paragraphs in any Indian language.