We build a two-pass Mandarin Automatic Speech Recognition (ASR) decoder on mobile device (PDA). The first-pass recognizing base syllable is implemented by discrete Hidden Markov Model (HMM) with time-synchronous, tree-lexicon Viterbi search. The second-pass dealing with language model, pronunciation lexicon and N-best syllable hypotheses from first-pass is implemented by Weighted Finite State Transducer (WFST). The best word sequence is obtained by shortest path algorithms over the composition result. This system limits the application in travel domain and it decouples the application of acoustic model and the application of language model into independent recognition passes. We report the real-time recognition performance performed on ASUS P565 with a 800MHz processor, 128MB RAM running Microsoft Window Mobile 6 operating system. The 26-hour TCC-300 speech data is used to train 151 acoustic model. The 3-minute speech data recorded by reading the travel-domain transcriptions is used as the testing set for evaluating the performances (syllable, character accuracies) and real-time factors on PC and on PDA. The trained bi-gram model with 3500-word from BTEC corpus is used in second-pass. In the first-pass, the best syllable accuracy is 38.8% given 30-best syllable hypotheses using continuous HMM and 26-dimension feature. Under the above syllable hypotheses and acoustic model, we obtain 27.6% character accuracy on PC after the second-pass.

Das Ziel dieser Arbeit ist die Formulierung eines Ansatzes zum maschinellen Lernen von Sprachmorphologie, in dem letztere als Zeichenkettentransformationen auf ganzen Wörtern, und nicht als Zerlegung von Wörtern in kleinere stukturelle Einheiten, modelliert wird. Der Beitrag besteht aus zwei wesentlichen Teilen: zum einen wird ein Rechenmodell formuliert, in dem morphologische Regeln als Funktionen auf Zeichenketten definiert sind. Solche Funktionen lassen sich leicht zu endlichen Transduktoren übersetzen, was eine solide algorithmische Grundlage für den Ansatz liefert. Zum anderen wird ein statistisches Modell für Graphen von Wortab\-leitungen eingeführt. Die Inferenz in diesem Modell erfolgt mithilfe des Monte Carlo Expectation Maximization-Algorithmus und die Erwartungswerte über Graphen werden durch einen Metropolis-Hastings-Sampler approximiert. Das Modell wird auf einer Reihe von praktischen Aufgaben evaluiert: Clustering flektierter Formen, Lernen von Lemmatisierung, Vorhersage von Wortart für unbekannte Wörter, sowie Generierung neuer Wörter.

This study is exploring the possibilities of automatic morphological analysis of L-verbs in the Palula language by the help from Finite-state technology and two-level morphology along with supervised machine learning. The type of machine learning used are neural Sequence to Sequence models. A morphological transducer is made with the Helsinki Finite-State Transducer Technology, HFST, toolkit covering the L-verbs of the Palula Language. Several Sequence to Sequence models are trained on sets of L-verbs along with morphological tagging annotation. One model is trained with a small amount of manually annotated data and four models are trained with different amounts of training examples generated by the Finite-State Transducer. The efficiency and accuracy of these methods are investigated. The Sequence to Sequence model trained on solely manually annotated data did not perform as well as the other models. A Sequence to Sequence model trained with training examples generated by the transducer performed the best recall, accuracy and F1-score, while the Finite-State Transducer performed the best precision score. / Denna studie undersöker möjligheterna för en automatisk morfologisk analys av L-verb i språket Palula med hjälp av finit tillståndsteknik och två-nivå-morfologi samt övervakad maskininlärning. Den typ av maskininlärning som används i studien är neurala Sekvens till Sekvens-modeller. En morfologisk transduktor är skapad med verktyget Helsinki Finite-State Transducer Technology, HFST, som täcker L-verben i Palula. Flera Sekvens till Sekvens-modeller tränas på set av L-verb med morfologisk taggningsannotation. En modell tränas på ett litet set av manuellt annoterade data och fyra modeller tränas på olika mängder träningsdata som genererats av den finita tillstånds-transduktorn. Effektiviteten och noggrannheten för dessa modeller undersöks. Sekvens till Sekvens-modellen som tränats med bara manuellt annoterade data presterade inte lika bra som de andra modellerna i studien. En Sekvens till Sekvens-modell tränad med träningsdata bestående av genereringar producerade av transduktorn gav bästa svarsfrekvens, noggrannhet och F1-poäng, medan den finita tillstånds-transduktorn gav bästa precision.

Within the context of natural language processing, a lemmatiser is one of the most important core technology modules that has to be developed for a particular language. A lemmatiser reduces words in a corpus to the corresponding lemmas of the words in the lexicon. A lemma is defined as the meaningful base form from which other more complex forms (i.e. variants) are derived. Before a lemmatiser can be developed for a specific language, the concept "lemma" as it applies to that specific language should first be defined clearly. This study concludes that, in Setswana, only stems (and not roots) can act independently as words; therefore, only stems should be accepted as lemmas in the context of automatic lemmatisation for Setswana. Five of the seven parts of speech in Setswana could be viewed as closed classes, which means that these classes are not extended by means of regular morphological processes. The two other parts of speech (nouns and verbs) require the implementation of alternation rules to determine the lemma. Such alternation rules were formalised in this study, for the purpose of development of a Setswana lemmatiser. The existing Setswana grammars were used as basis for these rules. Therewith the precision of the formalisation of these existing grammars to lemmatise Setswana words could be determined. The software developed by Van Noord (2002), FSA 6, is one of the best-known applications available for the development of finite state automata and transducers. Regular expressions based on the formalised morphological rules were used in FSA 6 to create finite state transducers. The code subsequently generated by FSA 6 was implemented in the lemmatiser. The metric that applies to the evaluation of the lemmatiser is precision. On a test corpus of 1 000 words, the lemmatiser obtained 70,92%. In another evaluation on 500 complex nouns and 500 complex verbs separately, the lemmatiser obtained 70,96% and 70,52% respectively. Expressed in numbers the precision on 500 complex and simplex nouns was 78,45% and on complex and simplex verbs 79,59%. The quantitative achievement only gives an indication of the relative precision of the grammars. Nevertheless, it did offer analysed data with which the grammars were evaluated qualitatively. The study concludes with an overview of how these results might be improved in the future. / Thesis (M.A. (African Languages))--North-West University, Potchefstroom Campus, 2006.

Within the context of natural language processing, a lemmatiser is one of the most important core technology modules that has to be developed for a particular language. A lemmatiser reduces words in a corpus to the corresponding lemmas of the words in the lexicon. A lemma is defined as the meaningful base form from which other more complex forms (i.e. variants) are derived. Before a lemmatiser can be developed for a specific language, the concept "lemma" as it applies to that specific language should first be defined clearly. This study concludes that, in Setswana, only stems (and not roots) can act independently as words; therefore, only stems should be accepted as lemmas in the context of automatic lemmatisation for Setswana. Five of the seven parts of speech in Setswana could be viewed as closed classes, which means that these classes are not extended by means of regular morphological processes. The two other parts of speech (nouns and verbs) require the implementation of alternation rules to determine the lemma. Such alternation rules were formalised in this study, for the purpose of development of a Setswana lemmatiser. The existing Setswana grammars were used as basis for these rules. Therewith the precision of the formalisation of these existing grammars to lemmatise Setswana words could be determined. The software developed by Van Noord (2002), FSA 6, is one of the best-known applications available for the development of finite state automata and transducers. Regular expressions based on the formalised morphological rules were used in FSA 6 to create finite state transducers. The code subsequently generated by FSA 6 was implemented in the lemmatiser. The metric that applies to the evaluation of the lemmatiser is precision. On a test corpus of 1 000 words, the lemmatiser obtained 70,92%. In another evaluation on 500 complex nouns and 500 complex verbs separately, the lemmatiser obtained 70,96% and 70,52% respectively. Expressed in numbers the precision on 500 complex and simplex nouns was 78,45% and on complex and simplex verbs 79,59%. The quantitative achievement only gives an indication of the relative precision of the grammars. Nevertheless, it did offer analysed data with which the grammars were evaluated qualitatively. The study concludes with an overview of how these results might be improved in the future. / Thesis (M.A. (African Languages))--North-West University, Potchefstroom Campus, 2006.

No description available.

Title: Processing of Turkic Languages Author: Sibel Ciddi Department: Institute of Formal and Applied Linguistics, Faculty of Mathematics and Physics, Charles University in Prague Supervisor: RNDr. Daniel Zeman, Ph.D. Abstract: This thesis presents several methods for the morpholog- ical processing of Turkic languages, such as Turkish, which pose a specific set of challenges for natural language processing. In order to alleviate the problems with lack of large language resources, it makes the data sets used for morphological processing and expansion of lex- icons publicly available for further use by researchers. Data sparsity, caused by highly productive and agglutinative morphology in Turkish, imposes difficulties in processing of Turkish text, especially for meth- ods using purely statistical natural language processing. Therefore, we evaluated a publicly available rule-based morphological analyzer, TRmorph, based on finite state methods and technologies. In order to enhance the efficiency of this analyzer, we worked on expansion of lexicons, by employing heuristics-based methods for the extraction of named entities and multi-word expressions. Furthermore, as a prepro- cessing step, we introduced a dictionary-based recognition method for tokenization of multi-word expressions. This method complements...

Automatic translation has seen tremendous progress in recent years, mainly thanks to statistical methods applied to large parallel corpora. Transductions represent a principled approach to modeling translation, but existing transduction classes are either not expressive enough to capture structural regularities between natural languages or too complex to support efficient statistical induction on a large scale. A common approach is to severely prune search over a relatively unrestricted space of transduction grammars. These restrictions are often applied at different stages in a pipeline, with the obvious drawback of committing to irrevocable decisions that should not have been made. In this thesis we will instead restrict the space of transduction grammars to a space that is less expressive, but can be efficiently searched. First, the class of linear transductions is defined and characterized. They are generated by linear transduction grammars, which represent the natural bilingual case of linear grammars, as well as the natural linear case of inversion transduction grammars (and higher order syntax-directed transduction grammars). They are recognized by zipper finite-state transducers, which are equivalent to finite-state automata with four tapes. By allowing this extra dimensionality, linear transductions can represent alignments that finite-state transductions cannot, and by keeping the mechanism free of auxiliary storage, they become much more efficient than inversion transductions. Secondly, we present an algorithm for parsing with linear transduction grammars that allows pruning. The pruning scheme imposes no restrictions a priori, but guides the search to potentially interesting parts of the search space in an informed and dynamic way. Being able to parse efficiently allows learning of stochastic linear transduction grammars through expectation maximization. All the above work would be for naught if linear transductions were too poor a reflection of the actual transduction between natural languages. We test this empirically by building systems based on the alignments imposed by the learned grammars. The conclusion is that stochastic linear inversion transduction grammars learned from observed data stand up well to the state of the art.

Esta tesis presenta diversas contribuciones en los campos de la clasificación automática de texto, traducción automática y traducción asistida por ordenador bajo el marco estadístico. En clasificación automática de texto, se propone una nueva aplicación llamada clasificación de texto bilingüe junto con una serie de modelos orientados a capturar dicha información bilingüe. Con tal fin se presentan dos aproximaciones a esta aplicación; la primera de ellas se basa en una asunción naive que contempla la independencia entre las dos lenguas involucradas, mientras que la segunda, más sofisticada, considera la existencia de una correlación entre palabras en diferentes lenguas. La primera aproximación dió lugar al desarrollo de cinco modelos basados en modelos de unigrama y modelos de n-gramas suavizados. Estos modelos fueron evaluados en tres tareas de complejidad creciente, siendo la más compleja de estas tareas analizada desde el punto de vista de un sistema de ayuda a la indexación de documentos. La segunda aproximación se caracteriza por modelos de traducción capaces de capturar correlación entre palabras en diferentes lenguas. En nuestro caso, el modelo de traducción elegido fue el modelo M1 junto con un modelo de unigramas. Este modelo fue evaluado en dos de las tareas más simples superando la aproximación naive, que asume la independencia entre palabras en differentes lenguas procedentes de textos bilingües. En traducción automática, los modelos estadísticos de traducción basados en palabras M1, M2 y HMM son extendidos bajo el marco de la modelización mediante mixturas, con el objetivo de definir modelos de traducción dependientes del contexto. Asimismo se extiende un algoritmo iterativo de búsqueda basado en programación dinámica, originalmente diseñado para el modelo M2, para el caso de mixturas de modelos M2. Este algoritmo de búsqueda n / Civera Saiz, J. (2008). Novel statistical approaches to text classification, machine translation and computer-assisted translation [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/2502 / Palancia

[ES] La detección de palabras clave (Keyword Spotting, en inglés), aplicada a documentos de texto manuscrito, tiene como objetivo recuperar los documentos, o partes de ellos, que sean relevantes para una cierta consulta (query, en inglés), indicada por el usuario, entre una gran colección de documentos. La temática ha recogido un gran interés en los últimos 20 años entre investigadores en Reconocimiento de Formas (Pattern Recognition), así como bibliotecas y archivos digitales. Esta tesis, en primer lugar, define el objetivo de la detección de palabras clave a partir de una perspectiva basada en la Teoría de la Decisión y una formulación probabilística adecuada. Más concretamente, la detección de palabras clave se presenta como un caso particular de Recuperación de la Información (Information Retrieval), donde el contenido de los documentos es desconocido, pero puede ser modelado mediante una distribución de probabilidad. Además, la tesis también demuestra que, bajo las distribuciones de probabilidad correctas, el marco de trabajo desarrollada conduce a la solución óptima del problema, según múltiples medidas de evaluación utilizadas tradicionalmente en el campo. Más tarde, se utilizan distintos modelos estadísticos para representar las distribuciones necesarias: Redes Neuronales Recurrentes o Modelos Ocultos de Markov. Los parámetros de estos son estimados a partir de datos de entrenamiento, y las respectivas distribuciones son representadas mediante Transductores de Estados Finitos con Pesos (Weighted Finite State Transducers). Con el objetivo de hacer que el marco de trabajo sea práctico en grandes colecciones de documentos, se presentan distintos algoritmos para construir índices de palabras a partir de modelos probabilísticos, basados tanto en un léxico cerrado como abierto. Estos índices son muy similares a los utilizados por los motores de búsqueda tradicionales. Además, se estudia la relación que hay entre la formulación probabilística presentada y otros métodos de gran influencia en el campo de la detección de palabras clave, destacando cuáles son las limitaciones de los segundos. Finalmente, todas la aportaciones se evalúan de forma experimental, no sólo utilizando pruebas académicas estándar, sino también en colecciones con decenas de miles de páginas provenientes de manuscritos históricos. Los resultados muestran que el marco de trabajo presentado permite construir sistemas de detección de palabras clave muy rápidos y precisos, con una sólida base teórica. / [CA] La detecció de paraules clau (Keyword Spotting, en anglès), aplicada a documents de text manuscrit, té com a objectiu recuperar els documents, o parts d'ells, que siguen rellevants per a una certa consulta (query, en anglès), indicada per l'usuari, dintre d'una gran col·lecció de documents. La temàtica ha recollit un gran interés en els últims 20 anys entre investigadors en Reconeixement de Formes (Pattern Recognition), així com biblioteques i arxius digitals. Aquesta tesi defineix l'objectiu de la detecció de paraules claus a partir d'una perspectiva basada en la Teoria de la Decisió i una formulació probabilística adequada. Més concretament, la detecció de paraules clau es presenta com un cas concret de Recuperació de la Informació (Information Retrieval), on el contingut dels documents és desconegut, però pot ser modelat mitjançant una distribució de probabilitat. A més, la tesi també demostra que, sota les distribucions de probabilitat correctes, el marc de treball desenvolupat condueix a la solució òptima del problema, segons diverses mesures d'avaluació utilitzades tradicionalment en el camp. Després, diferents models estadístics s'utilitzen per representar les distribucions necessàries: Xarxes Neuronal Recurrents i Models Ocults de Markov. Els paràmetres d'aquests són estimats a partir de dades d'entrenament, i les corresponents distribucions són representades mitjançant Transductors d'Estats Finits amb Pesos (Weighted Finite State Transducers). Amb l'objectiu de fer el marc de treball útil per a grans col·leccions de documents, es presenten distints algorismes per construir índexs de paraules a partir dels models probabilístics, tan basats en un lèxic tancat com en un obert. Aquests índexs són molt semblants als utilitzats per motors de cerca tradicionals. A més a més, s'estudia la relació que hi ha entre la formulació probabilística presentada i altres mètodes de gran influència en el camp de la detecció de paraules clau, destacant algunes limitacions dels segons. Finalment, totes les aportacions s'avaluen de forma experimental, no sols utilitzant proves acadèmics estàndard, sinó també en col·leccions amb desenes de milers de pàgines provinents de manuscrits històrics. Els resultats mostren que el marc de treball presentat permet construir sistemes de detecció de paraules clau molt acurats i ràpids, amb una sòlida base teòrica. / [EN] Keyword Spotting, applied to handwritten text documents, aims to retrieve the documents, or parts of them, that are relevant for a query, given by the user, within a large collection of documents. The topic has gained a large interest in the last 20 years among Pattern Recognition researchers, as well as digital libraries and archives. This thesis, first defines the goal of Keyword Spotting from a Decision Theory perspective. Then, the problem is tackled following a probabilistic formulation. More precisely, Keyword Spotting is presented as a particular instance of Information Retrieval, where the content of the documents is unknown, but can be modeled by a probability distribution. In addition, the thesis also proves that, under the correct probability distributions, the framework provides the optimal solution, under many of the evaluation measures traditionally used in the field. Later, different statistical models are used to represent the probability distribution over the content of the documents. These models, Hidden Markov Models or Recurrent Neural Networks, are estimated from training data, and the corresponding distributions over the transcripts of the images can be efficiently represented using Weighted Finite State Transducers. In order to make the framework practical for large collections of documents, this thesis presents several algorithms to build probabilistic word indexes, using both lexicon-based and lexicon-free models. These indexes are very similar to the ones used by traditional search engines. Furthermore, we study the relationship between the presented formulation and other seminal approaches in the field of Keyword Spotting, highlighting some limitations of the latter. Finally, all the contributions are evaluated experimentally, not only on standard academic benchmarks, but also on collections including tens of thousands of pages of historical manuscripts. The results show that the proposed framework and algorithms allow to build very accurate and very fast Keyword Spotting systems, with a solid underlying theory. / Puigcerver I Pérez, J. (2018). A Probabilistic Formulation of Keyword Spotting [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/116834