As to the size of rhyme groups, we also need to be careful of over-counting our evidence, as the number of links drawn between characters exponentially increases, the more rhyme words are found in a rhyme group. If there are only two words in a rhyme group, only one link will be added to the network; if there are three rhyme words, three links will be added. In the instance of four words, six links are added, and for five, ten will be added, etc.
Music Production and Promotion – Make Money Selling Beats
The problem is that rhyme words in large rhyme groups will seem to be perfectly integrated in the whole rhyme network, since the they are all interlinked with each other, even if they occur only once. This makes large rhyme groups very vulnerable to irregular rhymes. In order to cope with this problem, it is important to normalize the data, and to reduce the weight one adds to a link in the network in proportion to the size of the rhyme group. A simple way to do so is to divide each co-occurrence of two rhyme words by the size of the group of rhymes with which they rhyme.
In this normalization, the highest value for w AB is exactly 1 for the minimal group size of 2, and the weighted degree of each node the sum of the weight of all links of a node is equal to its occurrence as a rhyme word in the whole corpus. Apart from weighting the links in our network of rhyme connections, one can also weight the nodes. The latter value may be more interesting, since it reflects the more general distribution of potential rhyme words in the corpus of ancient Chinese poems.
If a given rhyme word, for example, occurs very often in a potential rhyme position, but rarely rhymes with other words, the few instances when they do rhymes should be treated with a certain suspicion. Moreover, as demonstrated later in this current study, there are algorithms for network partitioning that include information regarding the weight of the nodes in a network. Based on the preliminary thoughts discussed in the previous section, the rhyme network was reconstructed as follows: 1.
Links between two characters were drawn whenever they occurred in a group of rhyme words. The edge weights were further normalized as follows: 1. All co-occurring sections were stored in memory and only counted the first time they appeared. The concrete values for the co-occurrence of two characters in a rhyme group were normalized by applying Formula 1. The code was implemented in Python. The source code along with the input data and further instructions on how to replicate the network reconstruction, as well as further analyses described below, is provided along with the Supplementary Material accompanying this paper.
It comprises nodes and links between the nodes. A is based on a force-directed layout, and B shows the connected components of the network. The graphic was created with Cytoscape Smoot et al. The size of nodes is proportional to the node weight, as measured by the number of occurrences in the corpus. The width of edges is proportional to the edge weight, measured as the normalized number of co-occurrences of characters in different rhyme groups across different stanzas. In total, the network consists of 90 connected components, most of them consisting of less than five characters.
The fact that rhyme networks are almost completely connected comes as a bit of a surprise. Given the number of rhyme categories identified by the classical rhyme analysis, one might expect a clearer separation of at least the categories that seemed to be rather obvious to the classical scholars.
Yet when taking a closer look at the large connected component in the network, one can easily see that this component is itself structured, consisting of several groups which are more densely connected among themselves than with other groups.
Welcome to the Purdue OWL
When looking at this group in isolation, we can further see that most of the rhyme words form a connected component, with only a few outliers that do not occur in the biggest component of the network. This demonstrates that the structures postulated by the scholars can definitely be found in the network. However, in order to test and investigate these structures further, one needs to turn to more specific methods of network analysis. In network approaches, natural groups of similar objects are called communities. One should, of course, always be careful making analogies between different domains.
The analogy between social communities and rhyme groups, however, turns out to be quite fruitful. As mentioned above, classical rhyme analysis tries to determine groups of similar objects in the network of rhyme connections by searching for connected components. The analogy with communities, however, allows us to refine the strong claim of connected components. One no longer requires that all rhyme words connected in the network form a cluster of words with a similar pronunciation, but instead it can be proposed that the identified clusters should have more common edges among themselves than with other nodes outside the cluster.
The task of identifying rhyme categories in a network of rhyme patterns can thus be modeled as a community detection task. We still need to be careful regarding the analogy between communities and rhyme categories. Rhyme categories are usually thought to represent clusters of words with similar finals, and similarities are defined as commonalities between the nucleus and the coda of the words. When this is the case, however, it is by no means necessary that words with similar pronunciation in their finals actually rhyme in any collection of texts. However, it may well be possible that distinct communities discovered in this network still have the same pronunciation.
The reasons for this are manifold. It could be pure coincidence that connections are not made, but the semantics of rhyme words might also play a role, forcing groups of words from incompatible semantic or pragmatic domains to never rhyme with each other. One could think of similar social communities in different geographical locations here: the communities of football fans from Manchester and London are beyond doubt tightly connected in Manchester and London, but due to geographic distance and typical rivalries among football fans, it may be hard to find a football fan from Manchester who befriends a football fan from London in social networks.
In order to find the missing link between similar but separated communities, further data is required. The similarity between the Manchester and the London community of football fans could, for example, be shown by comparing the itineraries of the representatives, which may show that both regularly attend the same football stadiums. For our rhyme networks, we could include information regarding the phonetics of the characters, which can give us additional hints regarding their phonetic similarity.
This investigation would, however, go beyond the scope of this paper, and it is left for future research to pursue and test the fruitfulness of networks constructed from mixed data types. With help of algorithms for community detection, we can make this structure become transparent. For this purpose, an Infomap community detection analysis was carried out on the rhyme data. Infomap is Rosvall and Bergstrom is a fast community detection algorithm with a very good performance.
It handles weighted nodes and weighted edges, and uses random walks through the network in order to find the best partition of the network into communities. The Infomap analysis splits the network into distinct communities of varying size. This number is much higher than the 59 basic rhyme categories proposed by Baxter and Sagart This is, however, not surprising. Secondly, the network originally consisted of 90 connected components that community detection algorithms will automatically keep separate, since there is no evidence to connect the groups further.
In the figure, the category labels for these six categories were determined by taking the most frequently occurring rhyme in the reconstruction by Baxter and Sagartas representative of the whole group. The distribution of the six largest communities inferred by the Infomap algorithm over the network. The labels, following the reconstruction by Baxter and Sagart are determined in a majority-rules fashion by taking the most frequently occurring coda per community as the representative value of the whole set of characters. The results of this community detection analysis are available as an interactive web-based application.
The application displays all different codas accounted for in a given Infomap community. What is particularly surprising is the high- resolution power of the Infomap analysis. One needs to keep in mind, however, that the algorithm is solely based on the analysis of the rhyme patterns, and no additional evidence, be it Middle Chinese readings, the phonetics of the characters, or Sino-Xenic readings, was used.
Furthermore, as it was mentioned above, the realistic limits of our analogy between communities and rhyme categories need to taken into consideration. Not all words that show a similar rhyme behavior are necessarily also similar or identical in the pronunciation of their finals. The uniformity of the evidence sets the limits for community detection approaches applied to rhyme networks, but it remains a very useful starting point for both exploratory rhyme analysis, and for the testing of specific hypotheses.
How can specific hypotheses be tested or refined with help of the Infomap cluster analysis? This hypothesis goes originally back to a proposal by Starostin , and has been constantly gaining acceptance among researchers Hill , this volume. The test proposed in this paper is fairly simple. In order to make it easier to inspect the visual representation of this subnetwork, different colors are used to label the nodes. When investigating this community closer, however, it turns out that the mixed status is due to an artifact of the data. One of the specific features of the reconstruction system by Baxter and Sagart is that they are very explicit about uncertainties of their reconstruction.
In cases where evidence is not found to be sufficient to decide for one value only, they propose a tentative reconstruction value but put it in square brackets, thus making clear that they are not completely sure about the validity of the claim. The Infomap analysis not only justifies the uncertainty displayed in the Baxer and Sagart reconstruction, it also provides us with new suggestions regarding the reconstruction of the finals of this cluster.
This example shows that we can actively use the Infomap community analysis to review, test, and improve given reconstruction systems. A gives a force-directed view of the network while B shows the Infomap clusters. C and D show two different views of the Infomap clusters. This study presented a new way to approach the problem of rhyme analysis for the reconstruction of Old Chinese phonology. It was shown how the classical rhyme analysis, based on rhyme judgments applied to the Book of Odes can be used to construct a weighted network of rhyme words that was then further investigated through the standard approaches to network analysis.
Since the approach is still strictly experimental, no complete revision of current problems is now presented. Instead, the study attempted to demonstrate how strict network models of rhyme data can be used to test hypotheses in Old Chinese phonology and improve certain reconstruction proposals.
Several improvements need to be made in the future. The data needs to be enhanced, ideally incorporating additional analyses of rhyme patterns in the Book of Odes , similar to the one presented by Wang. Potentially many other studies should be added, although it is difficult to digitize the rhyme judgments in cases where the data is not presented transparently.
The meta-data also needs to be refined and completed, including the different available reconstructions of Old Chinese phonology. It would also be beneficial to incorporate alternative perspectives on the available data, especially the phonetic series. Ideally, all of the recent reconstruction proposals for Old Chinese should be digitally available as a series of rhyme judgments along with the proposed reconstruction for the rhyme words. This analysis presents two interactive applications that are supposed to ease the work of experts who often are not satisfied by just seeing the grand picture, but also wish to zoom in to reconstruct better a unique history of each word.
It is important to find ways to incorporate this incredibly valuable knowledge into the big data perspective, thus reconciling automatic and manual approaches to linguistic reconstruction. No matter whether one is accustomed to automatic approaches or not, it seems indispensable that historical linguists generally enhance the way they present their ideas to others, especially in those cases involving larger datasets.
In an ideal world, all the different ideas regarding the reconstruction of Old Chinese would be presented in a form that is both human- and machine-readable, thus enabling computational scientists to run large-scale analyses, while at the same time saving the experts invaluable time by providing quick access to the opinions of their colleagues. Network analyses are very common in many branches of science. Hopefully, this study represents starting point, and many more analyses of other aspects of Chinese historical linguistics that are amenable for network modeling will follow.
Here, it is less important to disprove the great work that has been done by classical Chinese linguists and current experts. I am very thankful to Laurent Sagart, William Baxter, Guillaume Jacques, and Nathan Hill, as well as the anonymous reviewers, who all commented on earlier versions of this manuscript and the network methods and helped me with their inspiring critics.
There is also no consensus regarding which final of the pair should be identified with the single final occurring after acute initials. Karlgren also sought to determine the phonetic values of the abstract categories yielded by the formal analysis, by comparing the categories of the Guangyun with other types of evidence, each of which presented their own problems. The Song dynasty rime tables applied a sophisticated featural analysis to the rime books, but were separated from them by centuries of sound change, and some of their categories are difficult to interpret. The so-called Sino-Xenic pronunciations, readings of Chinese loanwords in Vietnamese, Korean and Japanese, were ancient, but affected by the different phonological structures of those languages.
Finally modern varieties of Chinese provided a wealth of evidence, but often influenced each other as a result of a millennium of migration and political upheavals. After applying a variant of the comparative method in a subsidiary role to flesh out the rime dictionary evidence, Karlgren believed that he had reconstructed the speech of the Sui-Tang capital Chang'an. Later workers have refined Karlgren's reconstruction.
The initials of the Qieyun system are given below with their traditional names and approximate values: .
Websites Sold On Flippa.com
In most cases, the simpler inventories of initials of modern varieties of Chinese can be treated as varying developments of the Qieyun initials. The voicing distinction is retained in Wu Chinese dialects, but has disappeared from other varieties. Except in the Min Chinese dialects, a labiodental series has split from the labial series, a development already reflected in the Song dynasty rime tables. The retroflex and palatal sibilants had also merged by that time.
In Min dialects the retroflex stops have merged with the dental stops, while elsewhere they have merged with the retroflex sibilants. In the south these have also merged with the dental sibilants, but the distinction is maintained in most Mandarin Chinese dialects. The palatal series of modern Mandarin dialects, resulting from a merger of palatal allophones of dental sibilants and velars, is a much more recent development. Assigning phonetic values to the finals has proved more difficult, as many of the distinctions reflected in the Qieyun have been lost over time. However Pulleyblank, noting the use of these syllables in the transcription of foreign words without such a medial, claims the medial developed later.
Most reconstructions posit a large number of vowels to distinguish the many Qieyun rhyme classes that occur with some codas, but the number and the values assigned vary widely.
14 Music Production & Music Marketing eBooks Under $5
However the preface of the recovered Qieyun suggests that it represented a compromise between northern and southern reading pronunciations. The three groups are treated as tongyong in the Guangyun and have merged in all modern varieties. From early in the Tang dynasty, candidates in the imperial examination were required to compose poetry and rhymed prose in conformance with the rhyme categories of the Qieyun. However, the fine distinctions made by the Qieyun were found overly restrictive by poets, and Xu Jingzong and others suggested more relaxed rhyming rules.
It became the standard for official rhyme books, and was also used as the classification system for such reference works as the Peiwen Yunfu. Yan Zhengqing 's Yunhai jingyuan c. A side-effect of foreign rule of northern China between the 10th and 14th centuries was a weakening of many of the old traditions. The Zhongyuan Yinyun was a radical departure from the rhyme table tradition, with the entries grouped into 19 rhyme classes each identified by a pair of exemplary characters.
These rhyme classes combined rhymes from different tones, whose parallelism was implicit in the ordered of the Guangyun rhymes. The rhyme classes are subdivided by tone and then into groups of homophones, with no other indication of pronunciation. The dictionary reflects contemporaneous northern speech , with the even tone divided in upper and lower tones, and the loss of the Middle Chinese final stops.
Further innovations are found in a rime dictionary from the late 16th century describing the Fuzhou dialect , which is preserved, together with a later redaction, in the Qi Lin Bayin. This work enumerates the finals of the dialect, differentiated by both medial and rhyme, and classifies each homophone group uniquely by final, initial and tone. Both finals and initials are listed in ci poems. Tangut was the language of the Western Xia state — , centred on the area of modern Gansu.
The language had been extinct for four centuries when an extensive corpus of documents in the logographic Tangut script were discovered in the early 20th century. The dictionary consists of one volume each for the Tangut level and rising tones, with a third volume of "mixed category" characters, whose significance is unclear.
As with the Chinese dictionaries, each volume is divided into rhymes, and then into homophone groups separated by a small circle. The pronunciation of the first Tangut character in each homophone group is given by a fanqie formula using a pair of Tangut characters. Mikhail Sofronov applied Chen Li's method to these fanqie to construct the system of Tangut initials and finals.
From Wikipedia, the free encyclopedia. This article is about the type of dictionary in ancient China. For the type of Western reference work used in poetry, see Rhyming dictionary. I have taken the sounds and the rhymes of the various specialists and the dictionaries of the ancients and moderns, and by arranging what those before me have recorded, I have made up the five volumes of the Qieyun.
The splits and analyses are exceedingly fine and the distinctions abundant and profuse. Ramsey . Main article: Pingshui Yun. These final stops have disappeared in northern dialects, including the standard language, with the words distributed among the four modern tones. They are mostly alveolar in modern Chinese varieties. It is omitted in many reconstructions, and has no standard Chinese name. Chen Li was the first to realize in that they were distinguished in the Qieyun.
However in the Qieyun system [j] patterns with the palatals. Baxter, William H. Coblin, W. Creamer, Thomas B. Pulleyblank, Edwin G. Ramsey, S.
- Free verse.
- Festival Peak.
- A Guide to Internal Loss Prevention.
Dictionary Glossary Lexicon Phrase book Thesaurus. List of lexicographers List of online dictionaries.