Length N-grams Research Articles

Mining Deviations in Document Writing Style through Vector Dissimilarity

Doubts arise about the originality of a document when noticing a change in its writing style. This evidence to plagiarism has made the intrinsic approach for detecting plagiarism uncover the plagiarized passages through the analysis of the writing style for the suspicious document where a reference corpus to compare with is absent. The proposed work aims at discovering the deviations in document writing style through applying several steps: Firstly, the entire document is segmented into disjointed segments wherein each corresponds to a paragraph in the original document. For the entire document and for each segment, center vectors comprising average weight of their word are constructed. Second, the degree of closeness is calculated through applying Cosine similarity to measure for each segment, the deviation of its center vector from the center vector of the entire document. Additionally, word n-gram length will be investigated to show its effect on the proposed system performance wherein, center vectors are computed considering word n-grams for different values of n (n= 1, 2, and 3). Performance evaluation of the proposed method was accomplished through the use of Precision, Recall, F-measure, Granularity, and Plagdet as evaluation measures. Moreover, PAN-PC-09 and PAN-PC-11 were used for detecting intrinsic plagiarism as evaluation corpora. It is shown that the proposed approach has achieved results that are comparable to the state-of-the-art methods. Positive impact was observed through discovering deviations in document writing style by computing weight vectors dissimilarity rather than calculating the difference between the word n-grams that exist in segments and their corresponding word n-grams in the suspicious document. Furthermore, when considering the length of word n-gram, better results were recorded for system performance when word bi-grams was used compared to word uni-grams and word tri-grams.

Psycholinguistics of Organizational Phenomena: A Case of the Managerial Culture Study

Purpose. This article is devoted to the case study of relevant linguacultural stereotypes of the particular organization’s managerial culture and based on corresponding results the inquiry of the discourses formation features associated with the lexico-semantic meanings dispersion of (Foucault).
 Methods and Procedure of Research. Top managers of a large Ukrainian enterprise (67 respondents) were asked to arbitrarily describe the following concepts – “manager”, “subordinate”, “managerial style”. Each concept was differentiated according to the principle of the lexico-semantic opposition (“productive – counterproductive”). The obtained set of texts was lemmatized and a frequency analysis of the lemmas was carried out as well. Collocations were also evaluated, in particular, repeated n-grams were identified. In the further analysis were used all detected n-grams and those lemmas, the observed frequencies of which statistically significantly exceeded the expected ones.
 Results. Discourse formation features are determined by the paradigmatic relations of lexico-semantic units (words) and their semantic linear compatibility. The paradigmatic dispersion of a discourse is represented by the quality of connections between lemmas according to the criterion of the paradigm types to be found. Also informative is the assessment of the lemmas number (granularity of the discourse) and the spectrum diversity of paradigmatic relations in the discourse. These parameters reflect the properties of homogeneity/complexity of discourse meanings dispersion. The syntagmatic dissemination of discourse is manifested by the features of the words semantic linear interrelationship in each n-gram. This parameter reflects the quality of rationalizations (verbal-logical chains) associated with the corresponding discourses. The length of n-grams (the number of words) is also indicative – the higher n, the more differentiated the discourse is and vice versa. Matching the words semantics in n-grams with lemmas allows one to draw a conclusion regarding the internal consistency (integration) of the discourse.
 Conclusions. Psycholinguistic tools can be effectively used in an organizational phenomena study. In particular, the frequency analysis of lemmas, their lexico-semantic evaluation, as well as the qualitative analysis of n-grams in the problem-targeted texts of respondents allows one to solve applied research issues related to the assessment of current linguacultural trends in an organization and understanding their possible causes. Managerial culture, as a linguacultural phenomenon, is represented by a set of stably reproduced discourses in the process of the organization’s functioning. At the same time, discourses are constituted by the dispersion of meanings (words), which is being realized in two linguistic dimensions – paradigmatic and syntagmatic ones.

Zeta revisited

Abstract It has been claimed that traditional interpretations of Zeta based on graphs with bisector lines (Craig and Kinney, 2009, Shakespeare, Computers, and the Mystery of Authorship. Cambridge: Cambridge University Press.) are unsound, that validation with a held-out segments is inadequate, that counting types rather than tokens is suspect, and that the relationship between stronger Zeta scores and shorter word n-grams is an artifact of the method rather than a research finding (Rizvi, 2019a, The interpretation of zeta test results. Digital Scholarship in the Humanities, 34(2): 401–18). All of these claims are unsound. The separation of base and counter segments in a Zeta analysis is in fact an important research finding, the traditional interpretation of Zeta results based on a bisector line remain sound, and validation with a held-out segments is an appropriate method. Zeta’s reliance on consistency rather than frequency is not a bug but valuable feature that provides an important complementary method to those based on frequency. The results of extensive testing on corpora of prose, poetry, and drama show that increasing word n-gram length (but not character n-gram length) is often negatively correlated with classification accuracy. Higher Zeta scores are often correlated with better results, though the results vary a great deal depending on the corpus, the classification method, and the type and length of n-gram. Most importantly, these results show that Zeta gives strong results that are competitive with methods like Cosine Delta and Support Vector Machine on these classification tasks.

An Effective Entity Resolution Approach for Big Data

Entity Resolution (ER) is defined as the process 0f identifying records/ objects that correspond to real-world objects/ entities. To define a good ER approach, the schema of the data should be well-known. In addition, schema alignment of multiple datasets is not an easy task and may require either domain expert or ML algorithm to select which attributes to match. Schema agnostic meta-blocking tries to solve such a problem by considering each token as a blocking key regardless of the attributes it appears in. It may also be coupled with meta-blocking to reduce the number of false negatives. However, it requires the exact match of tokens which is very hard to occur in the actual datasets and it results in very low precision. To overcome such issues, we propose a novel and efficient ER approach for big data implemented in Apache Spark. The proposed approach is employed to avoid schema alignment as it treats the attributes as a bag of words and generates a set of n-grams which is transformed to vectors. The generated vectors are compared using a chosen similarity measure. The proposed approach is a generic one as it can accept all types of datasets. It consists of five consecutive sub-modules: 1) Dataset acquisition, 2) Dataset pre-processing, 3) Setting selection criteria, where all settings of the proposed approach are selected such as the used blocking key, the significant attributes, NLP techniques, ER threshold, and the used scenario of ER, 4) ER pipeline construction, and 5) Clustering where the similar records are grouped into the similar cluster. The ER pipeline could accept two types of attributes; the Weighted Attributes (WA) or the Compound Attributes (CA). In addition, it accepts all the settings selected in the fourth module. The pipeline consists of five phases. Phase 1) Generating the tokens composing the attributes. Phase 2) Generating n-grams of length n. Phase 3) Applying the hashing Text Frequency (TF) to convert each n-grams to a fixed-length feature vector. Phase 4) Applying Locality Sensitive Hashing (LSH), which maps similar input items to the same buckets with a higher probability than dissimilar input items. Phase 5) Classification of the objects to duplicates or not according to the calculated similarity between them. We introduced seven different scenarios as an input to the ER pipeline. To minimize the number of comparisons, we proposed the length filter which greatly contributes to improving the effectiveness of the proposed approach as it achieves the highest F-measure between the existing computational resources and scales well with the available working nodes. Three results have been revealed: 1) Using the CA in the different scenarios achieves better results than the single WA in terms of efficiency and effectiveness. 2) Scenario 3 and 4 Achieve the best performance time because using Soundex and Stemming contribute to reducing the performance time of the proposed approach. 3) Scenario 7 achieves the highest F-measure because by utilizing the length filter, we only compare records that are nearly within a pre-determined percentage of increase or decrease of string length. LSH is used to map the same inputs items to the buckets with a higher probability than dis-similar ones. It takes numHashTables as a parameter. Increasing the number of candidate pairs with the same numHashTables will reduce the accuracy of the model. Utilizing the length filter helps to minimize the number of candidates which in turn increases the accuracy of the approach.

N-Gram based language processing using Twitter dataset to identify COVID-19 patients

Due to the rapid growth of electronic documents, e.g., tweets, blogs, Facebook posts, snaps in different languages that use the same writing script, language categorization, and processing have great importance. For instance, to identify COVID-19 positive patients or people’s emotions on COVID-19 pandemic from tweets written in 35 different languages faster and accurate, language categorization and processing of tweets is significantly essential. Among many language categorization and processing techniques, character and word n-gram based techniques are very popular and simple but very efficient for categorizing and processing both short and large documents. One of the fundamental problems of language processing is the efficient use of memory space in implementing a technique so that a vast collection of documents can be easily categorized and processed. In this paper, we introduce a framework that categorizes the language of tweets using n-gram based language categorization technique and further processes the tweets using the machine-learning approach, Linear Support Vector Machine (LSVM), that may be able to identify COVID-19 positive patients. We evaluate and compare the performance of the proposed framework in terms of language categorization accuracy, precession, recall, and F-measure over n-gram length. The proposed framework is scalable as many other applications that involve extracting features and classifying languages collected from social media, and different types of networks may use this framework. This proposed framework, also being a part of health monitoring and improvement, tends to achieve the goal of having a sustainable society.

Predictive Model for Risk of 30-Day Rehospitalization Using a Natural Language Processing/Machine Learning Approach Among Medicare Patients with Heart Failure

IntroductionNearly 80% of all patients with heart failure (HF) are older adults (≥65 years of age). Prior studies have built predictive models that relied on structured data from electronic health records (EHRs) to predict the risk of 30-day rehospitalization for patients with HF. Structured data mostly included simple vocabularies such as age, and ethnicity. Rarely do prior studies include clinical narrative data in a free-text format (i.e., unstructured data). No previous study has focused on using clinical narrative notes specifically for Medicare patients with HF in the acute-care setting.AimTo identify clinical notes for building a predictive model for risk of 30-day rehospitalization among Medicate patients with HF.MethodsThis study first used free-text discharge summary notes and nursing care plans collected from June 1, 2015 to December 31, 2019, for a randomly selected 500 Medicare patients with HF. Natural Language Processing (NLP): we iterated over standard text pre-processing steps, exploring the impact of n-gram length, term document-frequency, word stemming, and the added value of parts-of-speech. We chose two models: 1) the classification model called Bag-of Words (BOW), where each document is represented by a vector based on the pre-processed text, and 2) Document Embedding, where document terms are mapped to a dimension-reducing layer (length equals 300). The latter runs exceptionally fast (on the order of tens-of-seconds for 2,000 documents). Machine Learning (ML): the output of the NLP BOW and Document Embedding models were fed to six different conventional machine learning systems (logistic regression, support vector machine, random forest, k-nearest neighbor clustering, a three-layer neural network, and Naïve Bayes).ResultsThe mean age was 77±7.9, and the average of length of hospital stay was 4.9 days ± 4.8. The best BOW model we found using discharge summaries (n=387) produced an Area Under the Receiver Operating Characteristics Curve (AUC) of 0.71 and F1 score of 0.65. The best Document Embedding model yielded an AUC of 0.65 and an F1 score of 0.61. Using nursing care notes as the unit of analysis (n = 2,046), the NLM/ML performed far better. The best BOW model on care plans found an AUC of 0.85 and F1 score of 0.77. The best Document Embedding produced an AUC of 0.83 and an F1 score of 0.75. In all cases we held out 33% of the data set for validation, repeating a random draw 10 times and averaging the performance results.ConclusionsWe conclude that nursing care plans are a better predictor of 30-day rehospitalization risk than discharge summaries. Because nursing care plans are shorter than discharge summaries, they have the added advantage of faster processing. Since the faster Document Embedding model's performance is comparable to that of BOW, we suggest its use in future work in the area of 30-day rehospitalization risk in Medicare patients with HF.

Abstract WP382: Using Natural Language Processing Algorithms to Identify Stroke Cases and Stroke Subtypes From Neuroimaging Reports

Background/Objective: The long-term goal of our research is to develop automated, accurate methods for conducting stroke outcome surveillance across large populations. To facilitate this, we developed natural language processing (NLP) based machine learning algorithms to classify stroke cases and sub-type strokes using neuroimaging reports. We report on the performance of our algorithms. Methods: Our population of interest included patients with stroke symptoms presenting to the emergency room of our large academic healthcare system. We randomly sampled 332 probable stroke cases. A trained neurologist validated stroke diagnoses using neuroimaging reports. Data preprocessing included cleaning and normalizing the reports into a standardized format. We trained and tested machine learning algorithms using the formatted reports. The NLP-based algorithms predicted a stroke diagnosis (binary classification) and a stroke type diagnosis (multiclass classification) using n-grams of length 1 (i.e., ‘stroke’, ‘hemorrhage’) through 3 (i.e., ‘no mass effect’), term-frequency weighting, and feature dimensionality reduction via truncated singular value decomposition (SVD). We report algorithm performance using the area under the receiver operating characteristic curve (AUC-ROC). Classification methods we tested included Multinomial Naïve Bayes, Logistic Regression, Random Forest Classifier, and Support Vector Machine (SVM). Results: The highest performing algorithm for both stroke and stroke sub-type classification contained 1 to 2 n-grams, no term-frequency, and 200 SVD components. For the stroke case detection, SVM achieved the best AUC-ROC of 95.6%. For stroke sub-type detection, SVM also yielded the highest AUC-ROC of 93.5% for no stroke, 92.3% for ischemic stroke, 91.9% for intraparenchymal hemorrhage, and 94.8% for subarachnoid hemorrhage. Conclusions: We report very promising results in our pilot study using machine learning algorithms to classify stroke cases from neuroimaging reports. Feature selection on this pilot data revealed a subset of words that are highly effective in categorizing stroke. Future work will focus on algorithm improvement, finer grained stroke sub-type stratification, and multi-label phenotyping.

Integrating Natural Language Processing and Machine Learning Algorithms to Categorize Oncologic Response in Radiology Reports.

A significant volume of medical data remains unstructured. Natural language processing (NLP) and machine learning (ML) techniques have shown to successfully extract insights from radiology reports. However, the codependent effects of NLP and ML in this context have not been well-studied. Between April 1, 2015 and November 1, 2016, 9418 cross-sectional abdomen/pelvis CT and MR examinations containing our internal structured reporting element for cancer were separated into four categories: Progression, Stable Disease, Improvement, or No Cancer. We combined each of three NLP techniques with five ML algorithms to predict the assigned label using the unstructured report text and compared the performance of each combination. The three NLP algorithms included term frequency-inverse document frequency (TF-IDF), term frequency weighting (TF), and 16-bit feature hashing. The ML algorithms included logistic regression (LR), random decision forest (RDF), one-vs-all support vector machine (SVM), one-vs-all Bayes point machine (BPM), and fully connected neural network (NN). The best-performing NLP model consisted of tokenized unigrams and bigrams with TF-IDF. Increasing N-gram length yielded little to no added benefit for most ML algorithms. With all parameters optimized, SVM had the best performance on the test dataset, with 90.6 average accuracy and F score of 0.813. The interplay between ML and NLP algorithms and their effect on interpretation accuracy is complex. The best accuracy is achieved when both algorithms are optimized concurrently.

An automatic classification of text documents based on correlative association of words

Training speed of the classifier without degrading its predictive capability is an important concern in text classification. Feature selection plays a key role in this context. It selects a subset of most informative words (terms) from the set of all words. The correlative association of words towards the classes increases an incertitude for the words to represent a class. The representative words of a class are either of positive or negative nature. The standard feature selection methods, viz. Mutual Information (MI), Information Gain (IG), Discriminating Feature Selection (DFS) and Chi Square (CHI), do not consider positive and negative nature of the words that affects the performance of the classifiers. To address this issue, this paper presents a novel feature selection method named Correlative Association Score (CAS). It combines the strength, mutual information, and strong association of the words to determine their positive and negative nature for a class. CAS selects a few (k) informative words from the set of all words (m). These informative words generate a set of N-grams of length 1-3. Finally, the standard Apriori algorithm ensembles the power of CAS and CHI to select the top most, b informative N-grams, where b is a number set by an empirical evaluation. Multinomial Naive Bayes (MNB) and Linear Support Vector Machine (LSVM) classifiers evaluate the performance of the selected N-Grams. Four standard text data sets, viz. Webkb, 20Newsgroup, Ohsumed10, and Ohsumed23 are used for experimental analysis. Two standard performance measures named Macro_F1 and Micro_F1 show a significant improvement in the results using proposed CAS method.

Machine learning techniques for XML (co-)clustering by structure-constrained phrases

A new method is proposed for clustering XML documents by structure-constrained phrases. It is implemented by three machine-learning approaches previously unexplored in the XML domain, namely non-negative matrix (tri-)factorization, co-clustering and automatic transactional clustering. A novel class of XML features approximately captures structure-constrained phrases as n-grams contextualized by root-to-leaf paths. Experiments over real-world benchmark XML corpora show that the effectiveness of the three approaches improves with contextualized n-grams of suitable length. This confirms the validity of the devised method from multiple clustering perspectives. Two approaches overcome in effectiveness several state-of-the-art competitors. The scalability of the three approaches is investigated, too.

Towards Learning Word Representation

Continuous vector representations, as a distributed representations for words have gained a lot of attention in Natural Language Processing (NLP) field. Although they are considered as valuable methods to model both semantic and syntactic features, they still may be improved. For instance, the open issue seems to be to develop different strategies to introduce the knowledge about the morphology of words. It is a core point in case of either dense languages where many rare words appear and texts which have numerous metaphors or similies. In this paper, we extend a recent approach to represent word information. The underlying idea of our technique is to present a word in form of a bag of syllable and letter n-grams. More specifically, we provide a vector representation for each extracted syllable-based and letter-based n-gram, and perform concatenation. Moreover, in contrast to the previous method, we accept n-grams of varied length n. Further various experiments, like tasks-word similarity ranking or sentiment analysis report our method is competitive with respect to other state-of-theart techniques and takes a step toward more informative word representation construction. Keywords: representation learning, n-gram model, NLP

A graph-based unsupervised N-gram filtration technique for automatic keyphrase extraction

In this paper, we present a novel N-gram (N> = 1) filtration technique for keyphrase extraction. To filter the sophisticated candidate keyphrases (N-grams), we introduce the combined use of: 1) statistical feature (obtained by using weighted betweenness centrality scores of words, which is generally used to identify the border nodes/edges in community detection techniques); 2) co-location strength (calculated by using nearest neighbour Dbpedia texts). We also introduce the use of N-gram (N> = 1) graph, which reduces the bias effect of lower length N-grams in the ranking process and preserves the semantics of words (phraseness), based upon local context. To capture the theme of the document and to reduce the effect of noisy terms in the ranking process, we apply an information theoretic framework for key-player detection on the proposed N-gram graph. Our experimental results show that the devised system performs better than the current state-of-the-art unsupervised systems and comparable/better than supervised systems.

Monolingual Information Retrieval using Terrier: FIRE 2010 Experiments Based on N-gram Indexing

Abstract N-gram based indexing technique has been proved as a useful technique for efficient document retrieval. We applied the n-gram approach and performed experiments in Hindi language text collections. The experiments are performed on the dataset of FIRE 2010 Hindi text collections. We used the Terrier open search engine for experimental purpose. Our experiments state that 4-gram gives the best results among all n-grams of different length. The results show an increase in value of mean average precision.

Web Page Genre Classification: Impact of n-Gram Lengths

pages are discriminated based on their topic and genre. Web page genres are capable to improve the modern search engines to focus on the user's information need. In this paper, web pages are represented using character n-grams. Character n-gram representation is language independent and allows automatic extraction of features from a web page. Character n-gram representation of a web page can be used efficiently to classify a web page by genre. Support Vector Machine (SVM) classification model is used for classification and experiments were carried out on 7-Genre corpus by varying the length of n-grams. It is observed that the performance in terms of F- measure improves as n-gram lengths are varied from 3 to 5 and it is also observed that performance degrades as the n- gram length is further increased.

A Hierarchical Approach to Makam Classification of Turkish Makam Music, Using Symbolic Data

A method for hierarchical classification of makams from symbolic data is presented. A makam generally implies a miscellany of rules for melodic composition using a given scale. Therefore, makam detection is to some level similar to the key detection problem. The proposed algorithm classifies makams by applying music theoretical knowledge and statistical evidence in a hierarchical manner. The makams using similar scales are first grouped together, and then identified in detail later. The first level of the hierarchical decision is based on statistical information provided by the n-gram likelihood of the symbolic sequences. A cross-entropy based metric, perplexity, is used to calculate similarity between makam models and the input music piece. Later, using statistical features related to the content of the piece, such as the tonic note, the average pitch level for local excerpts and the overall pitch progression, a more detailed identification of the makam is achieved. Different length n-grams and representation paradigms are used, including the Arel theory, the 12 tone equal tempered representation, and interval contour. Results show that the hierarchical approach is better, compared to a straightforward n-gram classification, for the makams which have similar pitch space, such as Hüseyni–Muhayyerand Rast–Mahur. Using the proposed methodology, the system’s recall rate increases from 88.7% to 90.9% where there exists still some confusion between the makams Uşşak and Beyati.

Comparing Medline citations using modified N-grams: Table 1

We aim to identify duplicate pairs of Medline citations, particularly when the documents are not identical but contain similar information. Duplicate pairs of citations are identified by comparing word n-grams in pairs of documents. N-grams are modified using two approaches which take account of the fact that the document may have been altered. These are: (1) deletion, an item in the n-gram is removed; and (2) substitution, an item in the n-gram is substituted with a similar term obtained from the Unified Medical Language System Metathesaurus. N-grams are also weighted using a score derived from a language model. Evaluation is carried out using a set of 520 Medline citation pairs, including a set of 260 manually verified duplicate pairs obtained from the Deja Vu database. The approach accurately detects duplicate Medline document pairs with an F1 measure score of 0.99. Allowing for word deletions and substitution improves performance. The best results are obtained by combining scores for n-grams of length 1-5 words. Results show that the detection of duplicate Medline citations can be improved by modifying n-grams and that high performance can also be obtained using only unigrams (F1=0.959), particularly when allowing for substitutions of alternative phrases.

Why so Similar?: Identifying Semantic Organizing Processes in Large Textual Corpora

This paper introduces the concept of semantic organizing processes as a means of inferring theoretically meaningful behavior from the observation of raw text. Semantic organizing processes are mechanisms by which a set of authors come to produce texts that are similar in some observable, quantifiable way. We introduce three broad semantic organizing processes -- authors sharing subject matter, authors sharing goals, and authors sharing sources -- and argue that each of these processes will lead to texts that tend to share n-grams at different lengths: short n-grams for shared subject matter, moderate length n-grams for shared goals, and long n-grams for shared sources. To test these hypotheses, we develop a novel n-gram extraction technique to capture text similarity based on n-grams of different lengths. We then apply our technique to a corpus where the author attributes are observable: the public statements of the Members of the U.S. Congress. Our results support the hypothesis that these three processes are reflected in distinct kinds of textual similarity. This article presents the first empirical finding that different social processes are detectable through the structure of overlapping textual features. The finding has important implications for modeling text and understanding underlying social processes.

A Copy Detection Method for Chinese Text by Character Based N-gram

This paper studies Chinese character-based n-grams used in copy detection by Sogou Chinese News Corpus. Texts were compared each other to find out the same news event written by different authors which were viewed as plagiarism by variable length (2<=n<=10) n-gram comparisons. The experiments show that unlike English, higher level (n={4, 5}) n-gram can improve precision significantly with no recall declined in Chinese text copy detection.

An Effective Method of Arbitrary Length N-gram Statistics for Chinese Text

N-gram frequency is always an important indicator in corpus-based natural language processing. As the corpus being larger, time and space consumption is always a serious issue. In this paper, we proposed a two-stage inverted index based on bigrams. The proposed index is constructed by two levels of variable length vectors and the two characters of a bigram are hashed to different levels. In order to ensure an efficient indexing, some null units are reserved in the vectors. By this way, unigrams and bigrams can be indexed directly. With the help of the index, the frequency of arbitrary length ngram can be calculated efficiently by intersection operation. The two-stage index has excellent properties: (1) it can be constructed efficiently; (2) the space consumption is significantly reduced with no compression; (3) when n=1 and n=2, time complexity of statistical processing decreases significantly. When n>2, it is more efficient than suffix array in most cases. This method is suitable for Chinese and some other Asian languages.

Quantitative evidence for a hypothesis regarding the attribution of early Buddhist translations

This article provides quantitative evidence for a hypothesis concerning fourth-century translations of Indian Buddhist texts from Prakrit and Sanskrit into Chinese. Using a Variable Length n-Gram Feature Extraction Algorithm, principal component analysis and average linkage clustering we are able to show that 24 sutras, attributed by the tradition to different translators, were in fact translated by the same translator or group of translators. Since part of our method is based on assigning weight to n-grams, the analysis is capable of yielding distinctive features, i.e. strings of Chinese characters, that are characteristic of the translator(s). This is the first time that these techniques have successfully been applied to medieval Chinese texts. The results of this study open up a number of new directions for the lexicographic and syntactic study of early Chinese translations of Buddhist texts.