No full-text available
To read the full-text of this research,
you can request a copy directly from the author.
Natural Language Processing: An Overview
Abstract
This paper is a lightly edited version of the slides for a talk of the same title. It provides a brief introduction to the subject matter of natural language processing, its relationship with artificial intelligence and its methodology. It includes slightly more detailed discussion of a pair of examples of exemplary natural language processing problems—speech recognition and machine translation.
... Sleutelwoorde: fonetiese spraaksegmentering, fonetiese belyning, spraaksintese, teks-na-spraak, spraakkorpusontwikkeling, hulpbron-arm tale, versteekte Markovmodelle, dinamiese tydverstelling. (Paulo and Oliveira, 2004 ...
... Given an audio recording of speech in a certain language, the task of phonetic annotation can be regarded as the combination of two sub-tasks (Paulo and Oliveira, 2004): ...
... • The definition of confidence measures, usually derived from information generated by the segmentation process itself (e.g. a DTW or HMM-based technique) (Paulo and Oliveira, 2004). ...
The rapid development of corpus-based speech systems such as concatenative synthesis systems for under-resourced languages requires an efficient, consistent and accurate solution with regard to phonetic speech segmentation. Manual development of phonetically annotated corpora is a time consuming and expensive process which suffers from challenges regarding consistency and reproducibility, while automation of this process has only been satisfactorily demonstrated on large corpora of a select few languages by employing techniques requiring extensive and specialised resources. In this work we considered the problem of phonetic segmentation in the context of developing small prototypical speech synthesis corpora for new under-resourced languages. This was done through an empirical evaluation of existing segmentation techniques on typical speech corpora in three South African languages. In this process, the performance of these techniques were characterised under different data conditions and the efficient application of these techniques were investigated in order to improve the accuracy of resulting phonetic alignments. We found that the application of baseline speaker-specific Hidden Markov Models results in relatively robust and accurate alignments even under extremely limited data conditions and demonstrated how such models can be developed and applied efficiently in this context. The result is segmentation of sufficient quality for synthesis applications, with the quality of alignments comparable to manual segmentation efforts in this context. Finally, possibilities for further automated refinement of phonetic alignments were investigated and an efficient corpus development strategy was proposed with suggestions for further work in this direction. Thesis (M.Ing. (Computer Engineering))--North-West University, Potchefstroom Campus, 2009.
... Another measure, duration-independent overlap rate [18] was employed to evaluate alignment accuracy. For this measure, 100 randomly selected word instances were manually time aligned across the corpus (50 within the segments that were used for MAP adaptation, and 50 in the segments that were used for evaluation only). ...
Initial work towards processing Afrikaans spoken lectures in a resource-scarce environment is presented. Two approaches to acoustic modeling for eventual alignment are compared: (a) using a well-trained target-language acoustic model and (b) using an acoustic model from another language, in this case American English. The authors show that while target-language acoustic models are preferable, similar performance can be achieved by repeatedly bootstrapping with the American English model, segmenting and then adapting or training new models using the segmented spoken lectures. The eventual systems perform quite well, aligning more than 90% of a selected set of target words successfully.
... When AI researchers began developing parsers for natural language in the early days, there was a strong feeling among some that parsing should be done without the use of grammar rules. These AI researchers (for example, [8], [9], [10], [12]) called for the development of "semantics" parsers. They envisioned such a parser would, upon reading a word, retrieve its meanings immediately and use them to interpret the sentence. ...
Many early AI "theories" about how the mind works were implemented in LISP and such works brought much excitement to the field of AI research in the 70's and 80's. LISP and AI were inseparable then. Somehow such works rarely appear in AI conferences anymore and LISP went out of fashion. Given this re-birth of a conference on LISP, this paper will discuss a new GOFAI theory on language implemented in LISP. Like the good old days, this paper is part AI, part LISP.
... However, these methods fail to take into account the seriousness of errors based on the relative distances between boundaries. A confidence measure proposed by Paulo and Oliveira [4] , considers segmentation results by determining an " Overlap Rate " for each phone in the reference data set, which can be expressed as a percentage and thus gives an objective measure of alignment performance, which takes into account differing phone lengths. We employ this measure of segmentation accuracy. ...
Improving the pronunciations of non-native language learners is an important task in a multilingual society. We focus on segmental aspects of pronunciation, and investigate the design of automated assistants that can be used to improve (1) the articulation of phones and (2) the production of tone. Initial experiments investigated phone productions in English by speakers of Nguni languages, and the tone of English speakers when producing speech in an Nguni language. Our initial results are promising, and point to the improvements that are required to develop a practical system.
Named Entity Recognition and Classification is a process of identification of proper nouns in the text and classification of those nouns into certain predefined categories like person name, location, organization, date, time etc. Named Entity Recognition and Classification in Kannada is an essential and challenging task. The aim of this work is to develop a novel model for Kannada Named Entity Recognition and Classification, based on conditional Random Fields. The paper discusses the various issues in developing the proposed model. The details of implementation and performance evaluation are discussed. The experiments are conducted on a training corpus of size 95,127 tokens and test corpus of 5,000 tokens. It is observed that the model works with Precision, Recall and F1-measure of 85%, 85% and 82% respectively.
Story understanding researchers have built computer programs that read and answer questions about simple stories. They have proposed a number of knowledge structures and processing mechanisms useful for this task.
2007: PRASA When doing research into or building systems involving spoken language, one invariably relies on relevantly annotated speech data for analysis and incorporation into such systems. The authors investigate methods and parameters for a baseline phonetic segmentation system on a few South African languages with the intention of determining how accurately they can apply basic methods and characterising typical deficiencies with the goal of defining further refinement strategies. An HMM-based system with a single mixture per triphone is found to work well, though the accurate segmentation of plosives remains a challenge. Suggestions for addressing this challenge are presented
this paper. Because a GPSG is so closely related to a CFG, it was thought that the well-known efficient parsing techniques for CFGs could be applied, with modifications, to GPSGs, and that GPSGs would therefore be computationally tractable. Recently, however, Ristad (1985) has shown that this is not the case, and that the unrestricted GPSG parsing problem is NP-complete (on the total problem size, viz. grammar plus input sentence) . Even before Ristad's result was known, workers in this field had found the practical problems caused by the interaction of FCRs, FSDs, and the propagation conventions difficult to surmount. Briscoe's comments (1986) are typical: "Finally, the concept of privileged feature, its interaction with feature specification defaults and the bi-directionality of the head feature convention are all so complex that it is debatable how much use they would be in a practical system (even if we did manage to implement them)" (p. 1); "The interaction of feature co-occurrence restrictions, feature specification defaults and feature propagation proved very hard to implement/understand" (p. 2)
ResearchGate has not been able to resolve any references for this publication.