Tibetan Corpus Research Articles

Improving Sequence-to-sequence Tibetan Speech Synthesis with Prosodic Information

There are about 6,000 languages worldwide, most of which are low-resource languages. Although the current speech synthesis (or text-to-speech, TTS) for major languages (e.g., Mandarin, English, French) has achieved good results, the voice quality of TTS for low-resource languages (e.g., Tibetan) still needs to be further improved. Because prosody plays a significant role in natural speech, the article proposes two sequence-to-sequence (seq2seq) Tibetan TTS models with prosodic information fusion to improve the voice quality of synthesized Tibetan speech. We first constructed a large-scale Tibetan corpus for seq2seq TTS. Then we designed a prosody generator to extract prosodic information from the Tibetan sentences. Finally, we trained two seq2seq Tibetan TTS models by fusing prosodic information, including feature-level and model-level prosodic information fusion. The experimental results showed that the proposed two seq2seq Tibetan TTS models, which fuse prosodic information, could effectively improve the voice quality of synthesized speech. Furthermore, the model-level prosodic information fusion only needs 60% ~ 70% of the training data to synthesize a voice similar to the baseline seq2seq Tibetan TTS. Therefore, the proposed prosodic information fusion methods can improve the voice quality of synthesized speech for low-resource languages.

Prompt-based for Low-Resource Tibetan Text Classification

Text classification is a critical and foundational task in Tibetan natural language processing, it plays a crucial role in various applications, such as sentiment analysis and information extraction. However, the limited availability of annotated data poses a significant challenge to Tibetan natural language processing. This paper proposes a prompt learning-based method for low-resource Tibetan text classification to overcome this challenge. This method utilizes pre-trained language models to learn text representation and generation capabilities on a large-scale unsupervised Tibetan corpus, enabling few-shot Tibetan text classification. Experimental results demonstrate that the proposed method significantly improves the performance of Tibetan text classification in low-resource scenarios. This work provides a new research idea and method for low-resource language processing, such as Tibetan natural language processing. Hopefully, it will inspire subsequent work on low-resource language processing.

Using Hybrid Penalty and Gated Linear Units to Improve Wasserstein Generative Adversarial Networks for Single-Channel Speech Enhancement

Recently, speech enhancement methods based on Generative Adversarial Networks have achieved good performance in time-domain noisy signals. However, the training of Generative Adversarial Networks has such problems as convergence difficulty, model collapse, etc. In this work, an end-to-end speech enhancement model based on Wasserstein Generative Adversarial Networks is proposed, and some improvements have been made in order to get faster convergence speed and better generated speech quality. Specifically, in the generator coding part, each convolution layer adopts different convolution kernel sizes to conduct convolution operations for obtaining speech coding information from multiple scales; a gated linear unit is introduced to alleviate the vanishing gradient problem with the increase of network depth; the gradient penalty of the discriminator is replaced with spectral normalization to accelerate the convergence rate of the model; a hybrid penalty term composed of L1 regularization and a scale-invariant signal-to-distortion ratio is introduced into the loss function of the generator to improve the quality of generated speech. The experimental results on both TIMIT corpus and Tibetan corpus show that the proposed model improves the speech quality significantly and accelerates the convergence speed of the model.

Problems of Semantic and Syntactic Modeling of Verbs in the Tibetan Corpus

Problems of Semantic and Syntactic Modeling of Verbs in the Tibetan Corpus

Cross-Language Transfer Learning-based Lhasa-Tibetan Speech Recognition

As one of Chinese minority languages, Tibetan speech recognition technology was not researched upon as extensively as Chinese and English were until recently. This, along with the relatively small Tibetan corpus, has resulted in an unsatisfying performance of Tibetan speech recognition based on an end-to-end model. This paper aims to achieve an accurate Tibetan speech recognition using a small amount of Tibetan training data. We demonstrate effective methods of Tibetan end-to-end speech recognition via cross-language transfer learning from three aspects: modeling unit selection, transfer learning method, and source language selection. Experimental results show that the Chinese-Tibetan multi-language learning method using multi-language character set as the modeling unit yields the best performance on Tibetan Character Error Rate (CER) at 27.3%, which is reduced by 26.1% compared to the language-specific model. And our method also achieves the 2.2% higher accuracy using less amount of data compared with the method using Tibetan multi-dialect transfer learning under the same model structure and data set.

Optimisation of the Largest Annotated Tibetan Corpus Combining Rule-based, Memory-based, and Deep-learning Methods

This article presents a pipeline that converts collections of Tibetan documents in plain text or XML into a fully segmented and POS-tagged corpus. We apply the pipeline to the large extent collection of the Buddhist Digital Resource Center. The semi-supervised methods presented here not only result in a new and improved version of the largest annotated Tibetan corpus to date, the integration of rule-based, memory-based, and neural-network methods also serves as a good example of how to overcome challenges of under-researched languages. The end-to-end accuracy of our entire automatic pipeline of 91.99% is high enough to make the resulting corpus a useful resource for both linguists and scholars of Tibetan studies.

An End-to-End Method for Joint Extraction of Tibetan Entity Relations

Entity relation extraction is to find entities and relations from unstructured texts, which is beneficial to the applications of knowledge graphs and question answering systems. The traditional methods handle this task in a pipelined manner which extracts the entities first and then recognizes their relations. This framework may lead to error delivery. In order to tackle this problem, this paper proposes an end-to-end method for joint extraction of Tibetan entity relations which can extract entities and relations at the same time. According to the Tibetan spelling characteristics, this paper processes the Tibetan corpus by word-level and character-level respectively. Combined with part of speech tagging, we use the end-to-end model to convert the entity relation extraction task to the tagging problem. Finally, the experimental results show that the proposed method is better than the baseline.

A Tibetan Language Model That Considers the Relationship Between Suffixes and Functional Words

The complete semantic representation of a Tibetan sentence is mainly determined by the addition of a specific functional word. The choice of Tibetan functional words is mainly influenced (both explicitly and implicitly) by the sequence of Tibetan suffixes. In this article, we propose an RNN-based Tibetan radical suffix unit (TRSU) to consider this relationship. Specifically, for the Tibetan radical suffix unit-explicit (TRSU-E) method, the fixed suffix in Tibetan is used to determine the virtual functional words. For the Tibetan radical suffix unit-implicit (TRSU-I) method, the decision is assisted by adding a specific suffix. To test the method, we design a standard Tibetan corpus, which consists of different genres. Our experimental results show that the complexity of our method is reduced by up to 22.2% relative to the best baseline. Furthermore, with the hidden semantic information and implicit suffix, TRSU-I outperforms TRSU-E by reducing the perplexity (PPL) by 3%. Moreover, good results are achieved on the English Penn Treebank data set.

Improved Distant Supervised Model in Tibetan Relation Extraction Using ELMo and Attention

The task of relation extraction is classifying the relations between two entities in a sentence. Distant supervision relation extraction can automatically align entities in texts based on Knowledge Base without labeled training data. For low-resource language relation extraction, such as Tibetan, the main problem is the lack of labeled training data. In this paper, we propose an improved distant supervised relation extraction model based on Piecewise Convolutional Neural Network (PCNN) to expand the Tibetan corpus. We add self-attention mechanism and soft-label method to decrease wrong labels, and use Embeddings from Language Models (ELMo) to solve the semantic ambiguity problem. Meanwhile, according to the Tibetan characteristics, we combine the word vector and part of speech vector to extract deeply feature of words. Finally, the experimental results show that P@avg value increases by 14.4% than baseline.

Особенности обработки тибетских композитов-существительных в лексической базе данных

Настоящая статья посвящена исследованию композитов-существительных корпуса текстов на современном тибетском языке с использованием реляционной лексической базы данных, содержащей единую непротиворечивую классификацию значений тибетских лексических единиц, с установлением между этими значениями различных семантических отношений. В статье описываются структура базы данных и принципы обработки тибетских композитов-существительных; типы семантических структур композитов, которые позволила выявить обработка. Поскольку большая часть композитов представляют собой грамматические, религиозно-философские и общенаучные термины, в статье анализируются особенности обработки концептов предметных областей знаний, в том числе, обусловленные буддийской языковой картиной мира.

Public opinion classification and text alignment based on Chinese and Tibetan corpus

To address the need for researching the security of Chinese and Tibetan networks, the classification of public opinion of Chinese and Tibetan texts is proposed. First, web pages are collected. Second, preprocessing is conducted to extract the useful information from web pages. Third, a table of the Chinese and Tibetan public opinion key words is built. Finally, text similarity calculation is proposed to classify the text according to the table of public opinion words. A Chinese–Tibetan text-level alignment approach that is based on Chinese and Tibetan translation dictionary is proposed to match word frequency and position. Furthermore, sentence-level alignment algorithm is studied. The alignment performance is related to the Chinese and Tibetan translation dictionary. Text classification of public opinion and Chinese–Tibetan text alignment system is developed. After public opinion classification of Chinese text, the alignment software can discover the most similar Tibetan text and present it to the user. This research can effectively contribute to identifying Chinese and Tibetan public opinion text and is meaningful for information retrieval, text clustering, and Chinese and Tibetan machine translation.

Research on Tibetan hot words, sensitive words tracking and public opinion classification

The rapid development of Tibetan information technology provides rich resources for Tibetan information processing technology. The construction of Tibetan corpus is the field of Tibetan information processing of basic work. In this paper, we design the system of Tibetan network data collection and web pages preprocessing. It can timely and efficiently access to web resources, and provide a basis for further analysis of Tibetan data. It can establish the Tibetan related corpus, enrich the Tibetan digital resources. It can also alleviate the status of Tibetan corpus data sparse and lack of resources and bring the convenient condition for Tibetan information processing. The hot words reflect the hot spot of Tibetan people’s attention in a certain period of time. Firstly, the paper proposes the method for reducing the space dimension of Tibetan news text. It can effectively reduce the complexity of subsequent processing. Secondly, term weighting method is proposed based on improved TFIDF for Tibetan text information extraction. It utilizes the idea that the words of different locations are given different weights to extract the hot words. On sensitive words discovery and classification of public opinion, sensitive thesaurus are collected artificially. Through the sensitive thesaurus comparison, the sensitive words are extracted. Classification of public opinion words is based on the proposed classification formula and the public opinion thesaurus. It will classify one Tibetan text to one public opinion class. In this paper, the software is developed to automatically collect Tibetan web pages from the network, preprocess the web pages, extract the text features and hot words, discover the sensitive words and classify the Tibetan text to one public opinion class. The experiment shows that the Tibetan hot words extraction is effective and Tibetan classification results of public opinion are significant.

Tibetan Person Attributes Extraction Based on Support Vector Machine

Person attributes extraction is an important branch in Natural Language Processing. At present, Chinese entity attributes extraction method is mainly based on rules, statistics and machine learning methods, and has achieved good results. But the study of Tibetan entity attributes extraction has a great space. The paper uses person attribute, case-auxiliary word, verbs and other related meaningful words in Tibetan corpus as the features, and utilizes support vector machine to train and predict the classifier. It provides support for search engine, information security, machine translation and other researches.

Semantic classification method for network Tibetan corpus

Tibetan web pages appear enormously. It is meaningful that the information processing technology is utilized to find the useful knowledge from the Tibetan web information. Tibetan semantic ontology can enrich the Tibetan digital resource and is helpful to improve the information processing performance. In this paper, semantic classification of Tibetan network corpus is studied. Firstly Tibetan web pages are collected. Secondly preprocessing is conducted to extract the useful information from Web pages. Thirdly the word segmentation and text representation are introduced. Finally the text similarity classification algorithm is proposed to classify the text. During the experiment, the comparison between semantic classification and non semantic classification is conducted. The results show that the semantic classification performance is obviously superior to non semantic classification. This means that making full use of ontology semantic relationship can greatly enhance the classification accuracy. The research is useful and helpful to the study of Tibetan semantic information processing.

Collection of Tibetan Network

With the development of Tibetan information technology, technologies about Tibetan web crawlers was extremely important. We elaborate different pages pretreatment rules according to the different sites and make the collected Tibetan Web text dump for Tibetan documents, by constructing a Web crawler to crawl different Tibetan websites, Experiments show that it can quickly and effectively to build large-scale Tibetan corpus, build the foundations for Tibetan information processing technology by self-made software and the module of pretreatment.

Tibetan Trisyllabic Light Verb Construction Recognition

The Tibetan trisyllabic light verb construction is a type of widely used verb phrase that is composed of a disyllabic noun or adjective and a light verb. A large number of Tibetan trisyllabic light verb constructions are widely found in Tibetan. Successfully recognizing this type of phrase greatly contributes to Tibetan information processing, however, thorough and systematic academic research in this field has not yet been launched. Therefore, we propose a model for the recognition of Tibetan trisyllabic light verb constructions based on an integrated strategy in this paper. Firstly, we extract all trisyllabic light verb construction candidates from a Tibetan corpus. In this step, light verbs are used as retrieval marks. Secondly, we filter candidates using a statistics-based model, rule-based model, and integrated model separately. Experimental results show that the integrated model performs much better than the other strategies, which proves that linguistic features contribute a lot to the automatic recognition of Tibetan trisyllabic light verb constructions by computers.

Research on Tibetan Semantic Role Labeling using an Integrated Strategy

Semantic role labeling is one of the most significant research fields of natural language processing. Researchers have already made many achievements in English and Chinese semantic role labeling. Until now, however, Tibetan semantic role labeling is still at an early stage due to the lack of a Tibetan corpus with semantic role annotation and relatively outdated research approaches. Tibetan is rich with syntactic markers that naturally divide a sentence into semantic chunks and indicate the semantic relationships between these chunks. Thus, in this paper, we propose a semantic role classification and an integrated strategy for Tibetan semantic role labeling. Transformation-Based Error-driven Learning and Conditional Random Fields have been employed in our study. Additionally, a number of linguistic rules have been introduced into our approach as well. Our integrated strategy achieves 83.91% in precision, 82.78% in recall, and an F-score of 85.71.

Research on the Extraction Technology of Hot-words in Tibetan WebPages

The construction of Tibetan corpus is the field of Tibetan information processing of basic work. This paper uses the technology of web crawler and pretreatment and real-time acquisition of web sites to obtain a large number of Tibetan corpus in short time. The hot words reflected the hotspot of Tibetan people’s attention in a certain period of time. The paper draws lessons from the TFIDF for Tibetan text information extraction and the words of different locations are given different weights to extract the hot words. It is really effective to realize the construction of the raw Tibetan corpus and the extraction of the hot-words by self-made software.

Design of Hot Issue Founding System Based on Tibetan Web

This paper designs a hot issue Founding system based on the news pages of Tibetan website, which can automatically find the issue on the network at any period of time. Through characteristics judgments of hot issues, and will present the conditions of issue meeting to the user. Because of large-scale corpus, and hot issues for demanding high real-time, then the system divided into groups according to days, and using agglomerative clustering to cluster daily corpus, and the corpus of the selected period of time using Single -pass clustering to get issue list, then order the issues through heat calculation. Finally, through testing Tibetan corpus in 2011, shows that the system has achieved good results, and with higher value.

The Phoneme Automatic Segmentation Algorithms Study of Tibetan Lhasa Words Continuous Speech Stream

In this paper, we adopt two methods to voice phoneme segmentation when building Tibetan corpus: One is the traditional artificial segmentation method, one is the automatic segmentation method based on the Mono prime HMM model. And experiments are performed to analyze the accuracy of both methods of segmentations. The results showed: Automatic segmentation method based tone prime HMM model helps to shorten the cycle of building Tibetan corpus, especially in building a large corpus segmentation and labeling a lot of time and manpower cost savings, and have greatly improved the accuracy and consistency of speech corpus annotation information.