This thesis compares the allocation tendency of sentential stresses among four reading styles, Lyric, Critical, Explanatory and Neutral. Three levels are investigated respectively named the chunk level, the base phrase level and the level of disyllabic prosodic words. Indicators for stress tendency are defined for clusters of words between chunks or within base phrases according to the research level to illustrate (1) the possibility for a class of words to obtain sentential stresses and (2) which component in a certain type of construction is easier to obtain stresses. The final conclusions are (1) rhythmic stresses tend to locate on the final part of the construction in levels of prosodic words and base phrases, regardless of reading styles; while in the chunk level, the stresses are generally evenly distributed. (2) allocation of semantic stresses is affected by the reading styles and varies in terms of the research level. Within prosodic words’, the initial-stressed cases are of absolute majority. In the chunk level, however, things are just the opposite. The final-stressed. tendency is dominant. Reading styles exert the biggest influence within base phrases. The Explanatory style shares a similar allocation tendency with the Neutral style. The Critical style differs from the Neutral style mainly in phrases with the adverbial+head structure. The Lyric style differs from the other styles in many aspects because, when reading essays in Lyric style, the speaker tends to form a kind of poetry-like rhythm to get .a better expression of the beauty of the essay.1. IntroductionStress, or its substitute, prominence, has been defined as "the degree of force" in terms of speech production [1] or as "the degree of loudness" from the viewpoint of speech perception [2]. It has been ranked into different levels of hierarchy, on the top of which is the most salient sentential stress [3]. The definition of sentential stress varies in past literatures. However, all these definitionscan be classified into two groups if the functions of the stress are considered the main factor in delivering messages. Generally speaking, those, as normal stress defined by Newman [4] and Zhao [5], or grammar stress defined by Bolinger [6] and Chomsky [7], to reflect the syntactic or rhythmic structure, are predictable with grammatical [8] or phonological rules [9]. Others, as contractive stress, emphatic stress [10], logical stress, to express speaker’s special intentions, are hard to predict without a deep understanding of the context.In Chu & Wang’s recent works [11][12], sentential stress is classified into rhythmic stress and semantic stress. The former serves the purpose of illustrating the rhythmic structure of an utterance and the later of making the speaker’s opinion or intention prominent. The validity of the classification has been proven by perceptual experiments. Their study shows that the rhythmic stress tends to be allocated to the final syllable within a prosodic word [12] and to the last syllable of the last word in a prosodic phrase [11], while no direct relationship has been found between the location of semantic stress and the prosodic structure of an utterance. In a study on semantic stress in Mandarin, Wang et. al. [12][13] found that the allocation tendency of semantic stress changes with the speech unit studied. On the one hand, semantic stress is more often allocated to the predicate part or the objective part (if there exists any) than to the subjective part in the chunk level,. However, such tendency does not hold within a base phrase. On the other hand, in a base phrase or a prosodic word, semantic stress is often found to be allocated to the modifiers when the phrase has an attribute+head structure or an adverbial+head structure. Yet, such tendency does not hold in the sentence level. The main difference in the allocation tendency of sentential stress between base phrases and prosodic words lies in that, semantic stress shows final-stress tendency in coordinative structures in base phrases but i