A Comparative Study of a Multi-Dimension/Multi-Feature Approach Between Chinese Debate and Speech

This study compares and analyzes Chinese debate and speech from the perspective of registers, which makes itself significant since it fills the gaps in the field. This paper is based on a self-built corpus and refers in Biber’s Multi-dimension/Multi-feature Approach. The quantitative statistical results show that there are 44 significant differences among the 65 linguistic features of debate and speech registers, and that these feature differences, from the macroscopic view, can be summarized into 8 different dimensions, which, to be specific, are known respectively as Multi-directional Interaction VS Single-directional Communication, Demonstrative VS Narrative, Intense Confrontation VS Deliberate Storytelling, Centralized Focuses VS Dispersing


Introduction
Great achievements in studying debate and speech have been made in China, yet few of them are done from the perspective of register. So far, I have found no monographs of register studies on debate or speech, with relevant articles less than 20. Nor have I seen any works comparing registers between debate and speech. Hence, this study is somewhat significant since it fills the gaps.
Quantitative methods have yet rarely been taken into account in terms of Chinese register studies, with a few involved going no further by merely investigating single or a few linguistic features. In the 1970s, some linguists abroad argued that similarities and differences among registers cannot be fully disclosed by analyzing individual linguistic features. Both Ervin-Tripp [1] and Hymes [2] insisted that "co-occurrence" should be used as a linguistic feature for register comparison. Yet confined to corpus and techniques at the time, it was not until the 1980s when Douglas Biber developed the multi-dimension/multi-feature approach that a better register comparison through multiple co-occurring linguistic features was accomplished.
Biber's multi-dimension/multi-feature approach is to identify several (usually dozens) linguistic features of different registers and have their frequencies compared statistically, which leads to a microscopic comparison among multiple features. Then these features would be generalized into several dimensions for macroscopic multi-dimensional investigation. And finally, functional explanations would be given according to situational characteristics. In his research process, Biber adopted qualitative methods for linguistic feature extraction, dimension identification and naming, as well as functional explanation in accordance with situational characteristics, and quantitative methods for judging differences among linguistic features, analyzing co-occurring factors and calculating dimension scores. In this way, Biber's multi-dimension/multi-feature approach would not bring about a biased result in analyzing registers as before when only single or a few linguistic features were focused, but also make itself a combination of qualitative and quantitative methods, macro-and micro-analysis, which helps to avoid the limitations caused by adopting a single method or analyzing from a single angle.
Biber proposed the multi-dimension/multi-feature approach in the 1980s and applied it into comparative studies of different registers for the following years [3][4][5][6][7][8][9][10][11]. Multi-dimension/multi-feature approach was introduced into China in the 2000s (see Lei Xiuyun [12] and Wu Jiangsheng [13,14]) and has been used since then for research, such as Ma Guanghui [15], Gui Shichun [16], Wen QiuFang [17], Hu Xianyao [18] etc. Most of them focused on English texts, though. Biber's multi-dimension approach was based on comparative analysis of English texts, which varies considerably from Chinese. Therefore, some linguistic features identified by Biber like derivation, that-clause, WH-clause with prepositional preposition etc. are not applicable for Chinese. Plus, investigation of multi-dimensions requires complicated statistical computations and software uses. As a result, few scholars in China have adopted this approach for register studies so far.
This paper attempts to comparatively study the registers between debate and speech via the multi-dimension/multi-feature approach, with emphasis on presenting an overall view of similarities and differences between debate and speech by comparing multiple linguistic features of them. The results could be used to improve apparent effects of register teaching in education, Chinese-English translation accuracy in translation, as well as operability of computer recognition for registers in information processing. Also, it could be assumed as a preliminary for multi-dimension studies.
The corpus used here was on the basis of the 4th Televised Speech Contest for Military University Students of China, which was themed with Military Revolution and included the speech session and the debate session. In details, recordings of 36 debaters in 6 debates, and speech recordings of 33 speakers were collected and transcribed. There is a total of 74019 Chinese characters (including punctuations) with debate 36536 and speech 37483, and of 46449 words with debate 21082 and speech 25367. Shi Xiaodong's Segmentation System for Chinese and SPSS 18.0 were adopted for word segmentation, comparison of means and independent samples t-test.

Comparison of Multiple Features Between Debate and Speech
Different situational characteristics and communicative purposes determined the frequencies of the same linguistic features in different registers. Through investigation and hypothesis testing of frequencies of linguistic features, a deep insight into similarities and differences of linguistic features among different registers is available. According to situational characteristics of debate and speech, along with communicative purposes, this paper identified 65 linguistic features and here are the results of quantitative comparisons.

Speed
Based on the fact that debate and speech are of different communicative purposes, in addition to our perception of reality, considerable differences between the speeds of them were expected. And a timing statistic was conducted for confirmation. The result showed that: they did bear much difference, with the average speed of debate reaching 303 characters per minute and speech 226 characters per minute 1 , while the normal speed would be 244 characters per minute 1This study tested the speed of the same persons in debate and speech (two people took part in speech and the opening and closing statement parts of debate) to eliminate bias caused by individual differences from statistical results. The result showed their speeds in speech are 226 characters per minute and 224 respectively, almost the same as the average speed in speech. And their speeds in the opening and closing statement in debate are 279 and 295, respectively, close to the average speed of opening and closing statement (which is 287 characters/ min). Thus, it is fair to consume that the statistic results of this study are unbiased and immune to individual differences. [19].

Register Markers and Common Words
Debate register has markers, such as my fellow debaters (对方辩友) and Thank you, chairman(谢谢主席), while speech doesn't have any. High-frequency words of debate are always related to reasoning, logistic, negation and stance, while those of speech are generally temporal and position nouns, toponyms or recollective words. Common words normally account for more than 10% of the two registers. For example, common words that are typical of debate register make up 11.40% of debate (the same words comprise merely 5.62% of speech); while common words that are typical of speech register make up 12.65% of speech (the same words comprise merely 4.72% of debate).  HF words refer to the top several high-frequency words in a text, with words sorted by descending order of frequency. Comparison of HF words helps to learn the concentration, states of content words and function words, nouns and pronouns et al. of different registers. The information would be further referred to for defining the dispersion of word uses, informativity and context-dependency between registers. The results show that the sum of tokens of top 20 HF words comprised 33.68% of the total in debate, and 28.13% in speech. It reveals that debate register has more high-frequency types and they are more concentrated. Besides, HF words of debate are composed of 75% content words, which are mostly nouns. The proportion of speech is 55% with pronouns in the majority. That is to say, the debate register is more informative than speech register, while speech is more dependent on contexts.

Rare Words
Rare words refer to low-frequency words, with a case in point being those that occur once and thus share a frequency of 1. The number of types or words that occur once is 3020 in debate, twice larger than speech. And in each frequency segment less than 30, speech has more types than debate; while in each segment over 30, debate has more types than speech. The fact that speech has more rare words and debate more high-frequency words makes types in speech nearly twice higher than that in debate, while the total tokens of the two are much the same. More rare words equals to a more dispersed use of types. That speech is more dispersed than debate in type uses confirms the statistical result of HF words above.

Coverage Ratio
According to Hou Min [20], coverage ratio is a percentage of specified subjects in the total amount of the examined corpus, namely the ratio of the cumulative sum of the frequency of each subject and its previous one to the total number of words of all subjects, with all subjects sorted by a descending order of frequency. The calculation formula is: ,nk is the frequency of subject k, N the total of all subjects, A the cumulative frequency and Ai is the cumulative frequency of A1 to Ai. This study took 500 words as a pitch and divided debate and speech into 5 subsections respectively of 500, 1000, 1500, 2000 and 2500 words. It is known from Illustration 1 that the coverage ratios of debate in all subsections are higher than speech. More HF words lead to higher recurrence ratio and coverage ratio of types. That debate has a higher coverage ratio than speech confirms the statistical result of HF words and rare words.

Normalized TTR
Type/token Ratio or TTR is generally used as an index for lexical diversity. However, it is confined to the text size, because a larger text brings about a lower TTR. So we had the texts normalized to acquire normalized TTR. The detailed operation was to compound the texts of the debate and speech into two big texts respectively and divided them into smaller ones with each containing 1000 words (including punctuations, and the word count of the last text is its actual amount if it's less than 1000). Then we got 22 small texts of debate and 26 of speech, and we calculated the TTR of each. Finally, we compared the mean values and independent samples t-test to find out if there are considerable differences of the normalized TTR between debate and speech. It could be seen that the average of normalized TTR of speech is higher than that of debate, and Sig. value of t-test is 0.000, which confirms that speech is significantly different from debate in normalized TTR and much more diverse lexically.

Lexical Density
There are various ways to get lexical density and that the number of content words (or lexical words) divided by the total number of words is usually adopted. As is often the case, a higher lexical density makes a text more informative. The statistical result show that debate bears a slightly higher lexical density than speech, yet there is no much difference between them.

Word Length & Sentence Length
In this section, the word length and the average word length of words with syllables ranging from 1 to 7, along with average length of complete sentences and clauses 2 were investigated. Previous studies found that a register of language, if more literate, has a higher share of disyllables; while if more oral, it has a higher share of monosyllables [21]. And literate registers have an advantage over oral registers in 2 A complete sentence is the sentence that ends with period, question mark, or exclamation mark, and a clause is the segment that ends with comma, semicolon or colon in a sentence. The average lengths of complete sentences and clauses refer to the average length of a complete sentence and that of a clause. terms of average word length and sentence length. The statistical result show that the proportion of disyllables in debate is slightly over that of monosyllables, meanwhile the situation in speech is to the contrary, in which the proportion of disyllables is much lower than that of monosyllables. The proportion of disyllables in debate is much higher than in speech, and the averages of word length, complete sentence and clause lengths are also slightly larger in debate. These linguistic features reflect an intensive literacy of debate register over speech.

Word-Class Distribution
30 items were identified for word-class investigation in accordance with the situational characteristics of debate and speech. Mean comparison and t-test saw significant differences of word-class distribution in 18 items, and none in the other 12. And word-class distributions of 9 classes are higher in debate than in speech and they are noun, verb, adverb, yuqi word, conjunction, interrogative pronoun, temporal word, particle, and "得" particle. Among them, the first 4 are of significant differences while the latter 5 are not. Besides, word-class distributions of 21 classes are higher in speech than in debate with 14 of them such as toponym, personal pronoun present significant differences, while the other 7 classes like person name and adjective don't. conjunction, interrogative pronoun, temporal word, particle, and "得" particle Higher Proportion in Speech over Debate Significant Differences preposition, position word, toponym, location word, personal pronoun, numeral, quantifier, "了", "地" and "着" particles, idiom, locution, lettered-word, state word Insignificant Differences pronoun, adjective, determiner, abbreviation, person name, onomatopoeia, "的" particle

Sentences
18 linguistic features like question sentence, complex sentence, negative, ellipsis and exclamation, parallelism and citation were investigated on the basis of situational characteristics and communicative purposes of debate and speech. It is obvious that debate has higher proportions of condition complex sentences, reason complex sentences, fanwen question 3 , shewen question 4 , WH-question, negation, citation, while speech has higher proportions of exclamation, metaphor or simile, analogy, parallelism and iteration. Also, debate has brief response and ellipsis that speech doesn't, and speech has metonym, antithesis, hyperbole and pun that debate doesn't.
Brief response and ellipsis in debate occur during the free debating stage. For instance, "Of course not." "OK." "Of course" "Thanks." "Certainly." "Definitely not." "Of course yes." "Good." "Great!" Brief response in debate differs from that in normal interactions, under which turn-taking usually takes place following the response. While in debate, more detailed explanations or demonstrations would be given before turn-taking. 3 A fanwen question is a figure of speech in the form of a question that is asked to express attitudes while requires no answer because it is already known. In the examples above, both B1 and D1 used ellipsis and brief response. The complete expression of B1 should be "Of course instruments are not the leaders of the band.", and that of D1 should be "Surely there are cases where the employment of certain technique length the war time." But the complete form seems to be too verbose and less powerful in free debating. As a result, debaters tend to use ellipsis and brief response, and explain or argue in details afterwards. It agrees with the specific situational characteristics and communicative purposes of debate. For instance: (1) 桂林的山太柔了，它承受不起巨大的伤痛；桂林的 水太弱了，它承载不住绵长的忧愁。 (2) 仰望蓝天，我豪情依旧；面对挑战，我壮志满怀。 在这场深刻的军事变革中，我将用理想、信念和忠诚标定 人生的坐标，用责任、素质和奉献托起腾飞的翅膀，实现 人生的第二次起飞！ (3) 我知道，只要霸权主义存在，阴云就会时时遮住和 平的阳光；我知道，如果我们的新军事变革不能持续快速 地推向前进，我们就难以积蓄起抗击强权的力量。 (4) 在他不大的办公室里，在堆积如山的资料和奖杯奖 状中间，从一幅他亲手书写的近代爱国诗人丘逢甲的《春 愁》里，你会找到答案："春愁难遣强看山，往事惊心泪 欲潸。四百万人同一哭，去年今日割台湾。" Translation: Mountains in Guilin are too supple to bear great pains, and its waters too fragile to take perpetual sorrows.
I am proud when looking up to sky and ambitious when faced with challenges. In this profound military revolution, I will calibrate the coordinates of life with ideal, faith, and loyalty, lift the soaring wings with responsibility, competency and dedication, and have my life had a second take-off.
I am aware that, if only the hegemony exists, the sunshine of peace would be overshadowed by dark clouds from time to time; and that if our new military revolution cannot be pushed forward constantly and rapidly, it would be hard for us to gather strength to fight against the great powers.
In his humble office, among those piled documents, trophies and award certificates, you could find the answer from a calligraphy work of his own that is titled with Spring Sorrow by the patriotic poet Qiu Fengjia: Spring sorrow is so overwhelming that I forced myself to enjoy the scenery; the past is too startling for me to hold back my tears. Four million people wept at the same time, for that Taiwan was occupied this day of last year.
Both (1) and (2) used antitheses in the first sentences, "第二 次起飞" (a second take-off) in example (2) is a pun, "阴云" (dark clouds) in (3) is a metonymy and "四百万人同一哭" (Four million people wept at the same time) in (4) is a hyperbole. Such figures of speech are missing in debate.

Multi-Dimension Differences Between Debate and Speech
Multi-feature analysis offers a thorough and comprehensive view of similarities and differences among different registers, yet which seem to take efforts to grasp overall with so many items and being too microscopic. Besides, some linguistic features generally co-occur as an indicator for a certain function. Factor analysis could help researchers to find out these co-occurred linguistic features and reduce the dimensions of register differences. Dimension reduction, to be exact, is decreasing dozens of linguistic features to several ones so that register differences could be better understood on the whole. Biber et al. [11], in their comparative studies of spoken and written registers in American universities, conducted dimension reduction by inducing the 90 linguistic features into 4 dimensions. So they had a macroscopic comparative study of registers. Another thing that should be noticed is that dimensional differences of registers are not static. When comparing different registers, Biber usually concluded several different dimensions on the basis of specific registers. Also, some linguistic features do not always occur in the same dimension. One single or several features occasionally co-occur in different dimensions and indicate different functions. 65 linguistic features of Chinese debate and speech were presented in the second part of this paper and 44 of them appear significant differences. On this basis, together with other insignificant differences, we summarized 8 dimensions of the register differences between debate and speech.

Multi-Directional Interaction VS Single-Directional Communication
Debate has much more verbs, adverbs, fanwen questions, shewen questions, and WH-questions than speech, as well as brief responses and ellipses, all of which are typical characteristics of interactive registers. Debate is interactive in nature. Its interactivity is beyond debaters and also embodied in the simple interaction between debaters and the chairman (like "OK, time's up!" "Thank you, chairman."), which makes debate multi-directional interactive. And speech is basically single-directional communicative, thus not equipped with the linguistic features of interactive registers.

Demonstrative VS Narrative
In Chinese speech, the frequencies of personal pronouns, "了" particle of past and perfect tense, and "着" particle of progressive tense are much higher than in debate. The result agrees with what Biber [6] wrote about linguistic features in narrative dimensions. Besides, the frequencies of prepositions, position words, location words, person names and toponyms in Chinese speech are higher than in debate. These features jointly indicate that speech is narrative. On the other hand, debate has a higher lexical density, and more nouns, yuqi words, conjunctions, reason complex sentences, condition complex sentences and citation. These features make debate demonstrative by being informative, explicitly expressive, intense in speaking and logic-emphasized, also citing extensively to strengthen its arguments.

Intense Confrontation VS Deliberate Storytelling
The speed of debate is quite faster than speech and the normal speed of speaking, which plus much more negations, fanwen question, shewen question, WH-questions and yuqi words, make debate confrontational. Negation, interrogating and questioning are generally used for confrontation, while higher speed, more verbs, adverbs and yuqi words have something to do with the inherent requirement of refutation and the pursuit of intense effects. On the contrary, speech has a slower speed, more pronouns, adjectives and person names, more "的"particle. And its prepositions, toponyms, location words, personal pronouns, metaphors or similes and exclamations are way too many over debate. Those agree with its communicative purposes to express thoughts or emotions, and arouse empathy by narrating and portraying. Only via deliberate storytelling with its reasonable narrative rhythm and slower overall speed, can the audience have a better understanding of the thoughts and emotions of the speaker.

Centralized Focuses VS Dispersing Contents
Lower normalized TTR, less word types and rare words, more HF words, and higher coverage ratio of word types indicate that debate is centralized in word type uses with higher recurrence ratio of same types, which is a reflection of centralized contents of debate register. In contrast, higher normalized TTR, more word types and rare words, less HF words, and lower coverage ratio of word types, along with lower recurrence ratio of same types indicate that speech is relatively dispersing in its content.

Precise Expressions VS Diverse Methods of Expressions
Debate, with argumentation being predominant, requires rigorous logics and explicit expression, while avoids ambiguity. That explains why debate has much more nouns, condition complex sentences, reason complex sentences and citations than speech and doesn't use figures of speech like metonymies or puns. The case is different in speech, though. To make their speeches more interesting and graphic, speakers usually make use of multiple methods: Firstly, lexical diversity (more rare words and higher normalized TTR), which is related to the content dispersion of speech, and to the selection and pursuit of lexical diversity of the speaker. Secondly, more dispersed distributions of different syllables, which for debate are mostly mono-, di-and tri-syllables, and for speech are composed of more quadri-and quinque-syllables that make words more variable. Thirdly, idioms, locutions, abbreviations, lettered-words, state words, onomatopoeias generally appear much more frequently in speech than in debate. Finally, speech has more exclamations, metaphors or similes, analogies, parallelism, and recurrences, along with more metonymies, antithesis, hyperboles and puns that don't show up in debate.

Informative VS Affective
Debate, with content words taking a high share of its HF words, more nouns and conjunctions, higher lexical density, and larger averages of word lengths, complete sentence lengths and clause lengths, are more informative than speech. Speech, with its high proportions of adjectives, exclamations, parallelism, recurrence, hyperbole, is relatively more affective.

Specialized VS Universal
Terms occurred frequently in debate, with quite a few just in HF words, like war, military, technology, information, let alone in the whole texts, such as asymmetry, nonlinearity, operation modes, military technologies, military structure, leading elements, system engineering, and mechanized revolution. These terms make the debate register specialized in a way. Though "military revolution" is also the topic of speech, such military terms barely occurred in speech. Speakers basically tell stories that happen in daily life and use daily vocabularies, which make the language of speech universal to the public.

Literate Style VS Oral Style
Debate adopts many terminologies to make it more literate, which is also confirmed by its slightly higher proportion of disyllables over monosyllables. The proportions of monosyllables and quantifiers are higher in oral style registers than literate style registers, see Liu Yanchun [21][22][23].In speech, the proportion of monosyllables is much higher than that of disyllables, so are the proportions of monosyllables and quantifiers over debate. That makes speech more of oral style or colloquial than debate, which is also confirmed by the fact that the proportion of disyllables is much higher in debate, plus that the lexical density, averages of word lengths, complete sentence lengths and clause lengths of debate are slightly higher over speech.

Functional Explanations for Differences of Linguistic Features and Dimensions between Debate and Speech
Biber et al. [11] believed that linguistic features, instead of being arbitrary, are in accordance with the specific situation of different registers. Hence, only the combination of linguistic features and situational characteristics can basic dimensions of registry varieties be better explained and described. Biber offered a frame for analysis of situational characteristics, which is universally applicable for any registers to analyze the situational characteristics, or to be exact, to analyze the relationship between participants, channels, production circumstance, settings, communicative purposes and topics. In terms of these items, analyses of how linguistic features and dimensional differences of Chinese debate and speech correspond with the situational characteristics are given below.

Participants
This item could be investigated from the aspects of the addressor and the addressee. In speech, the addressor is a single person, and the addressee is a group of people, who may not be provided with the involving knowledge. It requires the speech to be less professional in the contents and the language so that audiences could understand and accept the speaker's thoughts and emotions as much as possible. That's why dictions of speech are basically from daily life, hardly involving terms. And that gives the speech the feature of being universal. However, in the case of debate, the addressor and the addressee are mostly debaters of both sides (the chairman and judges also need to learn about the proposition in advance), who are prepared and ready for arguments to defense the viewpoints of his side. That's to say, they don't need to avoid being professional. Instead, more professionally a debater behave, more persuasive his arguments is. That explains the occurrences of terms in debate, which gives the debate the feature of being specialized.

Relationships Between Participants
Judging from the relationship between participants, speech is basically away from interactive (it is occasionally interactive, yet the interactions are pretty brief and merely used to arouse emotions or attract attentions of the audience), and debate is highly interactive, especially during the free debating phase. So there are some typical interactive linguistic features in the debate register, like addressing, responding and ellipsis etc. Also, because the two parties of debate have confrontation against their opinions, debate has a high speed and many negative words and sentences, fanwen and shewen questions. In speech's case, the speaker tends to slow down and tell stories deliberately so audiences would understand his thoughts and emotions.

Channels
Both debate and speech are in spoken forms, so the averages of word lengths, complete sentence length, and clause lengths of them are relatively smaller on the whole, and both of them have the feature of being colloquial. However, debate is a little more of literate style than speech because of its rigorous logics, standard languages, and large numbers of terms, conjunctions, logic words and disyllables, along with its slightly higher lexical density over speech.

Production Circumstance
Though in spoken forms, debate and speech can be well prepared like being planned and written in advance. That's why repetition, pleonasm or pauses etc. that are common in spoken language are missing from both. And they are both careful with dictions and strict with structures. Besides, to make the speech more variable and attractive, the speaker would employ multiple techniques of expressions and figures of speech, for example, controlling the recurrence ratio of words, using extensive rare words, collating different syllables, using idioms, locutions, abbreviations, lettered-words, onomatopoeias as well as metaphors or smiles, analogies, antitheses, hyperboles and puns and so on.

Setting
The setting refers to the physical context of the communication -the time and place. Though both sides of the debate, speaker and audience of the speech share the time and space, the contents of debate and speech actually are not necessarily related to the scene. Speech, with its contents focusing on recollection and reconstruction of the past, has many temporal words, toponyms, and position words that related to the past, along with recollective words, as well as more "了" particle of past and perfect tenses, and "着" particle of progressive tense over debate, all of which makes speech more narrative. And debate, with its contents focusing on the proposition, has more related terms and logic word relevant to ways of argumentation, and thus appear to be more demonstrative.

Communicative Purposes
As debate aims to support the viewpoints of their own side and refute the opposite, it has large numbers of register markers like "my fellow debaters", nouns, conjunctions, reason complex sentences, condition complex sentences, citations, and words that used to express standpoints and logical relations, and it appears to be highly informative and demonstrative. As speech aims to express thoughts and emotions, and arouse empathies through reconstructions of stories and imitation of objects, it has many related personal pronouns, person names, toponyms, position words, particles of "了" and "着" and appears to be highly narrative. Besides, large quantities of adjectives, state words, metaphors or similes, exclamations, parallelisms, and recurrences also make the speech more affective.

Topics
Topic is an open-ended category that can be described at many different levels. It is possible to distinguish among very general topical domains, such as science, religion, politics, and sports, but any text will have its own specific topics. In this paper, debate is generally about confrontational argumentations between two sides on a proposition and emphasizes thorough analyses and demonstrations on the argument, which in linguistic features are large amounts of nouns, terms, HF words, conjunctions and logic words, small amounts of rare words, along with low normalized TTRs. The situation is different in speech. Despite the same thememilitary as debate, speech is mostly about storytelling, with its contents focusing on touching stories about causes and lives. And because stories differ from speaker to speaker, speech covers content of wide ranges, which in linguistic features are small amounts of HF words, large amounts of rare words and types, high normalized TTRs, and dispersed information.

Conclusion
This paper attempts to adopt Biber's multi-dimension/multi-feature approach for a comparative study of Chinese debate and speech. In terms of linguistic features, instead of indiscriminately taking Biber's in his register studies, we identified 65 linguistic features for investigation on the basis of specific situational characteristics and communicative purposes of Chinese debate and speech, along with reality perception of differences between the two. It is in accordance with what Biber proposed as "the situational characteristics are more basic" and "registers are determined by their situational characteristics". What's more, speed in phonetic level, some figures of speech in rhetoric level plus sentence types are also included here as features. It is an extension of Biber's multi-features analyses, in which only vocabularies, characteristics of words, grammatical features and syntactic structures were investigated. In terms of practice methods for multi-feature investigation, except for some that are not testable, most linguistic features were tested with comparison of means and hypothesis testing, which basically implemented the objective identification of significant differences. However, in terms of multi-dimension analysis, the volume of the text corpus was insufficient for factor analysis, which caused failure of the objective recognition of identifying co-occurrences of linguistic features. Namely, this study didn't fully achieved what Biber advocated as quantified dimensional analysis. This is a tentative application of Biber's multi-dimension/multi-feature approach; it gives a deeper understanding of the application conditions and the statistical techniques this approach requires, and lays the foundation for further applying it into overall comparative studies of Chinese registers.

Notes
Work on this project was supported by National Social Science Foundation Research Grant #13CYY038 and Excellent Young Teacher Training Project Grant # YXJS201510. And this paper is a translation (translated by Qiu Le) of its Chinese version with a few changes in the title and abstract. The original one was published in the No. 6 (2016) issue of Language Teaching and Linguistic Studies. Afterwards the author further conducted a multi-dimensional study on 17 Chinese registers with 1112 texts and identified 72 linguistic features, which finally produced 7 primary dimensions by using factor analysis. However, some linguistic features identified in this study such as terms, HF words and rare words etc. could not make it into the 72, and some others that were not identified here like private verbs, public verbs did. As a result, this study is not consistent with its follow-up in specific dimensional induction. But they do have consistency in features between debate and speech in many co-occurred or similar dimensions. For example, speech had an advantage over debate with regard to narrative features, while debate turned to be more prominent in terms of interactivity and logicality. It thus ensures the trustworthiness in the dimensional analysis results of this paper, despite its failure to conduct factor analysis.