Improving word coverage using unsupervised morphological analyser

Powerful computers are needed for processing tasks related to human languages these days. Human languages, also called natural languages, are highly versatile systems of encoding information and can capture information of various domains. To enable a computer to process information in human languages, the language needs to be appropriately ‘described’ to the computer, i.e. the language needs to be ‘modelled’. In this work, we present an approach for acquisition of morphology of inflectional language like Hindi. It is an unsupervised learning approach, suitable for languages with a rich concatenative morphology. Broadly, our work is carried out in three steps: 1. Acquire the morphology of Hindi from a raw (un annotated) Central Institute of Indian Languages (CIIL), Mysore text corpus, 2. prepare clusters and prepare stem bag and suffix bag, 3. use the morphological knowledge to decompose given word as stems and suffixes according to their morphological behaviour and add new words. A prime motivation behind this work is to eventually develop an unsupervised morphological analyser which is language-independent (used for Hindi). Second motivation is to develop a Morphological segmentation which is language-independent as it is shown that study of morphology would benefit to a range of NLP tasks such as speech recognition, speech synthesis, machine translation and information retrieval. Though Hindi is an important and a national language in India, little computational work has been done so far in this direction. Our work is one of the first efforts in this regard and can be considered pioneering. There are many such languages for which it is very important to have a suitable but inexpensive computational acquisition process. Languages receive very little attention of computational linguistic research both in terms of availability of funds and number of researchers. We however do not claim that our approach is a solution for all such languages. Different languages have characteristics that require individual research attention.     

Spoken Indian language identification: a review of features and databases

Spoken language is one of the distinctive characteristics of the human race. Spoken language processing is a branch of computer science that plays an important role in human–computer interaction (HCI), which has made remarkable advancement in the last two decades. This paper reviews and summarizes the acoustic, phonetic and prosody features that have been used for spoken language identification specifically for Indian languages. In addition, we also review the speech databases, which are already available for Indian languages and can be used for the purposes of spoken language identification.     

Error analysis to improve the speech recognition accuracy on Telugu language

Speech is one of the most important communication channels among the people. Speech Recognition occupies a prominent place in communication between the humans and machine. Several factors affect the accuracy of the speech recognition system. Much effort was involved to increase the accuracy of the speech recognition system, still erroneous output is generating in current speech recognition systems. Telugu language is one of the most widely spoken south Indian languages. In the proposed Telugu speech recognition system, errors obtained from decoder are analysed to improve the performance of the speech recognition system. Static pronunciation dictionary plays a key role in the speech recognition accuracy. Modification should be performed in the dictionary, which is used in the decoder of the speech recognition system. This modification reduces the number of the confusion pairs which improves the performance of the speech recognition system. Language model scores are also varied with this modification. Hit rate is considerably increased during this modification and false alarms have been changing during the modification of the pronunciation dictionary. Variations are observed in different error measures such as F-measures, error-rate and Word Error Rate (WER) by application of the proposed method.     

A simple and efficient optical character recognition system for basic symbols in printed Kannada text

Optical Character Recognition (OCR) systems have been effectively developed for the recognition of printed characters of non-Indian languages. Efforts are on the way for the development of efficient OCR systems for Indian languages, especially for Kannada, a popular South Indian language. We present in this paper an OCR system developed for the recognition of basic characters (vowels and consonants) in printed Kannada text, which can handle different font sizes and font types. Hu’s invariant moments and Zernike moments that have been progressively used in pattern recognition are used in our system to extract the features of printed Kannada characters. Neural classifiers have been effectively used for the classification of characters based on moment features. An encouraging recognition rate of 96.8% has been obtained. The system methodology can be extended for the recognition of other south Indian languages, especially for Telugu.     

Reducing barriers of communication across Indian languages: Anai andes approach to mass media

The present paper identifies some nonlinguistic and linguistic barriers that will have to be overcome by any system for automatic and simultaneous communication of news, commercial advertisements, and other items of information and entertainment by mass media across some Indian languages. The paper also presents a brief account of some theories for the representation of knowledge of language in a language-independent manner, because such theories can make simultaneous communication of an item across different languages rather easy. But more research is required in this field before the relevant knowledge can be represented in a language-independent manner. Presently automatic and simultaneous communication of an item from any language to many other languages does not seem easy. However, it seems that in a very limited way a beginning can be made in the direction of such communication by human editors aided by tools developed by computer scientists so far. Revised and expanded version of a paper presented at the discussion meeting on “Artificial intelligence and expert system technologies in the Indian context” held at Indian Institute of Science, Bangalore, India, July 22–26, 1991 However, I alone am responsible for mistakes, if any, here.     

Automatic transcription of continuous speech into syllable-like units for Indian languages

The focus of this paper is to automatically segment and label continuous speech signal into syllable-like units for Indian languages. In this approach, the continuous speech signal is first automatically segmented into syllable-like units using group delay based algorithm. Similar syllable segments are then grouped together using an unsupervised and incremental training (UIT) technique. Isolated style HMM models are generated for each of the clusters during training. During testing, the speech signal is segmented into syllable-like units which are then tested against the HMMs obtained during training. This results in a syllable recognition performance of 42·6% and 39·94% for Tamil and Telugu. A new feature extraction technique that uses features extracted from multiple frame sizes and frame rates during both training and testing is explored for the syllable recognition task. This results in a recognition performance of 48·7% and 45·36%, for Tamil and Telugu respectively. The performance of segmentation followed by labelling is superior to that of a flat start syllable recogniser (27·8%and 28·8%for Tamil and Telugu respectively).     

基于声学特征的母语非汉语者声调研究

从提高满足少数民族普通话高自然度语音合成与高精度语音识别的实际应用需求出发,首次从实验语音学的角度对初级、中级和高级阶段的50名维吾尔族汉语学习者与10名母语为汉语普通话的说话人声调的一阶差分与时长以及相似度进行对比,并对其声调的一阶差分模式、声调时长等韵律参数进行了实验分析,得出维吾尔族学生对汉语声调的偏误情况以及与中国少数民族汉语水平等级考试(Master of Human Kinetics, MHK)成绩的关系。通过实验结果可以发现,三组维吾尔族人学习普通话的声调都有困难。两种语言的音系,语调和重音等特性影响了第二语言中的声调特性。归纳了维吾尔族学习者声调的基本声学特征,总结出了一些重要的规则和结论;为解决给汉语语音处理带来的困难,尤其是少数民族汉语的语音合成和语音识别方面的声调问题,提供了重要的参考依据。     

Role of neural network models for developing speech systems

This paper discusses the application of neural networks for developing different speech systems. Prosodic parameters of speech at syllable level depend on positional, contextual and phonological features of the syllables. In this paper, neural networks are explored to model the prosodic parameters of the syllables from their positional, contextual and phonological features. The prosodic parameters considered in this work are duration and sequence of pitch (F ₀) values of the syllables. These prosody models are further examined for applications such as text to speech synthesis, speech recognition, speaker recognition and language identification. Neural network models in voice conversion system are explored for capturing the mapping functions between source and target speakers at source, system and prosodic levels. We have also used neural network models for characterizing the emotions present in speech. For identification of dialects in Hindi, neural network models are used to capture the dialect specific information from spectral and prosodic features of speech.     

Variability of the southwest Indian Ocean

The variability in the southwest Indian Ocean is connected to the basin-scale and global-scale ocean circulation. Two bands of enhanced variability stretch across the Southern Indian Ocean east of Madagascar around 12 degrees S and 25 degrees S, respectively. They mark the preferred routes along which anomalies, generated by varying forcing over the central basin, near the eastern boundary or in the equatorial region, propagate westward as baroclinic Rossby waves. Sea-surface height anomalies pass along the northern tip of Madagascar and are observed by satellite altimetry to propagate into the central Mozambique Channel. There, eddies are subsequently formed that propagate southward into the Agulhas retroflection region. The anomalies along the southern band trigger the formation of large dipolar vortex pairs in the separation region of the East Madagascar Current at the southern tip of the island. South of Africa these eddies and dipoles trigger the shedding of Agulhas Rings that feed the Atlantic meridional overturning circulation with warm, salty, Indian Ocean water. Interannual variability of the forcing over the Indian Ocean, such as that associated with the Indian Ocean Dipole/El Nino climate modes, propagates along these pathways and leads to associated modulations of the eddy transports into the South Atlantic.     

Effect of pre-treatments on hydrolysis and methane production potentials of by-products from meat-processing industry

Language issues are problems with communication via speech, signs, gestures or their written equivalents. They may result from poor reading and writing skills, a mix of foreign languages and other circumstances. Language issues are not picked up as a safety risk on the shop floor by current safety management systems. These safety risks need to be identified, acknowledged, quantified and prioritised in order to allow risk reducing measures to be taken. This study investigates the nature of language issues related danger in literature, by experiment and by a survey among the Seveso II companies in the Netherlands. Based on human error frequencies, and on the contents of accident investigation reports, the risks associated with language issues were ranked. Accident investigation method causal factor categories were found not to be sufficiently representative for the type and magnitude of these risks. Readability of safety related documents used by the companies was investigated and found to be poor in many cases. Interviews among regulators and a survey among Seveso II companies were used to identify the gap between the language issue related dangers found in literature and current best practices. This study demonstrates by means of triangulation with different investigative methods that language issue related risks are indeed underestimated. A recommended coarse of action in order to arrive at appropriate measures is presented.     

Scaling and universality in the human voice

Speech is a distinctive complex feature of human capabilities. In order to understand the physics underlying speech production, in this work, we empirically analyse the statistics of large human speech datasets ranging several languages. We first show that during speech, the energy is unevenly released and power-law distributed, reporting a universal robust Gutenberg–Richter-like law in speech. We further show that such ‘earthquakes in speech’ show temporal correlations, as the interevent statistics are again power-law distributed. As this feature takes place in the intraphoneme range, we conjecture that the process responsible for this complex phenomenon is not cognitive, but it resides in the physiological (mechanical) mechanisms of speech production. Moreover, we show that these waiting time distributions are scale invariant under a renormalization group transformation, suggesting that the process of speech generation is indeed operating close to a critical point. These results are put in contrast with current paradigms in speech processing, which point towards low dimensional deterministic chaos as the origin of nonlinear traits in speech fluctuations. As these latter fluctuations are indeed the aspects that humanize synthetic speech, these findings may have an impact in future speech synthesis technologies. Results are robust and independent of the communication language or the number of speakers, pointing towards a universal pattern and yet another hint of complexity in human speech.     

Synthesis of unlimited speech in Indian languages using formant-based rules

Synthesis of continuous and unlimited speech is a matter of theoretical as well as technological interest. Independent efforts are needed for synthesis in Indian languages which are substantially different from English and other European languages. The paper discusses basic synthesis issues like text-to-phoneme and phoneme-to-speech conversion and incorporation of prosody. The three commonly adopted methodologies of concatenation, formant and articulatory syntheses are compared. The TIFR phoneme-to-speech synthesizer which utilizes a standard formant synthesizer as a speech production model is described and the methodology for evolving and organizing formant-based rules to drive the used synthesizer is emphasized. The results of some perception tests are reported and a few potential applications are suggested. The direction of the future work for enhancing the quality and expanding the scope of the synthesizer is indicated. Deceased     

Infrastructure for domain-specific aspect languages: the relax case study

Domain-specific aspect languages (DSALs) bring the well-known advantages of domain specificity to the level of aspect code. However, DSALs incur the significant cost of implementing or extending a language processor or weaver. Furthermore, this weaver typically operates blindly, making detection of interactions with aspects written in other languages impossible. This raises the necessity of an appropriate infrastructure for DSALs. The case study we present here illustrates how the Reflex kernel for multi-language AOP addresses these issues, by considering the implementation of a DSAL for advanced transaction management, KALA. We first detail the implementation of KALA in Reflex, called ReLAx, illustrating the ease of implementation of runtime semantics, syntax, and language translation. We then show a straightforward and modular extension to KALA at all these levels, and demonstrate how Reflex helps in dealing with interactions between KALA and another DSAL for concurrency management. These invaluable assets enable faster development of DSALs as well as their ability to coexist within one application, thereby removing the most important impediments to their re-emergence in the aspect community.     

Effective Persuasion of International Business Sales Letters

• Language usage has attracted increasing research attention in international business studies. Yet scarce research has been done on the use of language in sales letters across cultures.
• Our paper, using a combined etic-emic approach aims to compare Chinese and New Zealand managers’ reflective accounts of persuasive strategies and the function of sales letters used in these two countries. In particular, we seek to contribute to an improved in-depth cross-cultural understanding through an investigation using emic sources of language and persuasion.
• Our findings indicate that language and persuasion play a significant role in sales letters, the Chinese managers focusing more on building qing (positive affect) with the reader whereas the NZ managers sought a more immediate reaction to their sales pitch and informal engagement with the reader.
• Furthermore, also based on the insights gained reflectively and reflexively from our discussion, these differences were closely related to persuasion strategies which are also important parameters influencing cross-cultural adaptations.

     

信号相位匹配算法的语音降噪及性能评估

针对信号相位匹配算法能够在信号估计中准确提取信号频率成分的优点,将三子阵信号相位匹配算法应用于提取强背景噪声中的语音信号,同时提出了基于平均段内功率谱密度距离的语音质量客观评价方法,将主观平均意见分（MOS）、信噪比和平均分段功率谱密度距离作为指标,分别对三子阵信号相位匹配法、谱减法和最小均方误差估计法这三种算法的处理结果进行客观评价并对算法的降噪性能做了对比。仿真结果表明,三子阵相位匹配算法能够达到强背景噪声环境下语音降噪的目的;同时基于平均段内功率谱密度距离的语音质量客观评价结果与主观评价结果相符合,该评价方法具有一定的可行性。     

Perceiving and remembering speech depend on multifractal nonlinearity in movements producing and exploring speech

Speech perception and memory for speech require active engagement. Gestural theories have emphasized mainly the effect of speaker''s movements on speech perception. They fail to address the effects of listener movement, focusing on communication as a boundary condition constraining movement among interlocutors. The present work attempts to break new ground by using multifractal geometry of physical movement as a common currency for supporting both sides of the speaker–listener dyads. Participants self-paced their listening to a narrative, after which they completed a test of memory querying their narrative comprehension and their ability to recognize words from the story. The multifractal evidence of nonlinear interactions across timescales predicted the fluency of speech perception. Self-pacing movements that enabled listeners to control the presentation of speech sounds constituted a rich exploratory process. The multifractal nonlinearity of this exploration supported several aspects of memory for the perceived spoken language. These findings extend the role of multifractal geometry in the speaker''s movements to the narrative case of speech perception. In addition to posing novel basic research questions, these findings make a compelling case for calibrating multifractal structure in text-to-speech synthesizers for better perception and memory of speech.     

Nearest neighbour classification of Indian sign language gestures using kinect camera

People with speech disabilities communicate in sign language and therefore have trouble in mingling with the able-bodied. There is a need for an interpretation system which could act as a bridge between them and those who do not know their sign language. A functional unobtrusive Indian sign language recognition system was implemented and tested on real world data. A vocabulary of 140 symbols was collected using 18 subjects, totalling 5041 images. The vocabulary consisted mostly of two-handed signs which were drawn from a wide repertoire of words of technical and daily-use origins. The system was implemented using Microsoft Kinect which enables surrounding light conditions and object colour to have negligible effect on the efficiency of the system. The system proposes a method for a novel, low-cost and easy-to-use application, for Indian Sign Language recognition, using the Microsoft Kinect camera. In the fingerspelling category of our dataset, we achieved above 90% recognition rates for 13 signs and 100% recognition for 3 signs with overall 16 distinct alphabets (A, B, D, E, F, G, H, K, P, R, T, U, W, X, Y, Z) recognised with an average accuracy rate of 90.68%.     

Relative occurrences and difference of extrema for detection of transitions between broad phonetic classes

Detection of transitions between broad phonetic classes in a speech signal has applications such as landmark detection and segmentation. The proposed hierarchical method detects silence to non-silence transitions, sonorant to non-sonorant transitions and vice-versa. The subset of the extrema (minimum or maximum amplitude samples) above a threshold, occurring between every pair of successive zero-crossings, is selected from each frame of the bandpass-filtered speech signal. Locations of the first and the last extrema lie on either side far away from the mid-point (reference) of a frame, if the speech signal belongs to a non-transition segment; else, one of these locations lies within a few samples from the reference, indicating a transition frame. The transitions are detected from the entire TIMIT database for clean speech and 93.6% of them are within a tolerance of 20 ms from the phone boundaries. Sonorant, unvoiced non-sonorant and silence classes and their respective onsets are detected with an accuracy of about 83.5% for the same tolerance with respect to the labelled TIMIT database as reference. The results are as good as, and in some aspects better than, the state-of-the-art methods for similar tasks. The proposed method is also tested on the test set of the TIMIT database for robustness with respect to white, babble and Schroeder noise, and about 90% of the transitions are detected within a tolerance of 20 ms at the signal to noise ratio of 5 dB. On NTIMIT database, 62.7% of the transitions are detected, and 63.5% of the sonorant onsets, within 20 ms tolerance.     

Cosine histogram analysis for spectral image data classification

Conventional multivariate strategies for making qualitative estimates of sample composition rely chiefly on identifying subtle differences in spectral shape. In some instances, such as in biological tissues, the spectra obtained from a single sample class may consist of many shapes. Likewise, two distinctly different sample classes, such as normal and abnormal tissue, may produce similar variations in spectral shape. In our work, we employ statistical analysis of the set of cosine correlation scores obtained from multispectral visible absorption images of stained cervical Papanicolaou samples. By analyzing the cosine correlation score frequency for spectra obtained from the cell nuclei, abnormal cells can be differentiated from the background of normal cells, which vary considerably in their optical properties and morphology. Our method, called cosine histogram analysis (CHA), returns the percent likelihood of abnormality for each pixel in the field of view and is presented here for the first time.     

A Simulation Generator for Flexible Manufacturing Systems Design and Control

Recent developments in simulation languages have enabled the modeling of complex and large systems. Systems modeling using simulation has many advantages. Even though the creation of realistic models using a simulation language is becoming easier, it is not trivial. Simulation languages are not designed to be problem-specific. Furthermore, the art of systems modeling can be quite time consuming.

This paper describes a user-oriented simulation generator for the design and control of flexible manufacturing systems (FMS). The simulation generator serves as a preprocessor to the SIM AN simulation language. The generator converts data into a simulation model and automatically runs the simulation program. Standard output is provided by the simulation program, as well as plots and histograms of performance measures of interest to the analyst.

Model development is significantly simplified by the simulation generator described in this paper. Simulation generators can be developed in a relatively short period of time by using an existing simulation language. Simulation generators represent a very attractive alternative to specialized simulation languages.