A methodology for qualitative analysis of historical corpus data: New insights about language change from a micro-analytical approach

语言学家一直对找出语言是如何变化的感兴趣。这一研究领域在1980年代发生了变化，当时有大量历史数据(语料库)的电子资源可供研究。语料库第一次使用定量方法研究语言变化成为可能。历史语料库语言学一直致力于开发更多更好的语料库和量化方法。定性分析的方法尚未调整，以利用新资源提供的所有可能性。该项目的主要目标是纠正这种平衡，并将定性研究和方法提上议程。为此，它将提出新的定性方法来识别历史语料库数据中的变化，展示它们提供的描述性收益，表明用这种方法发现的新数据可以与历史语言学中长期存在的问题有关，例如语言变化是如何发生的。首先，我将开发新的方法来识别一种特定类型的变化，即语言项通过这种变化获得新的功能。一个例子是“确定的”一词的变化，它最初只出现在诸如“某一胜利”之类的组合中，其中它的意思是“确定的，有保障的”，其功能是描述所讨论的胜利的性质。通过语言上的改变，‘某些’发展了一种新的限定语功能：例如，‘某些人不喜欢猫’，而‘某些’相当于‘一些’。在这一变化中，“确定”的形式保持不变。这就提出了一个问题，我们如何在历史语料库数据中区分这两种功能，以及我们可以为这种变化援引哪些经验证据。新的方法将包括与物品的形式和/或物品(其分布)所处的背景范围有关的各种证据，因为在比较不同时期的数据时可以观察到这些方面的变化。因为这些方法调查单个项目，所以它们被认为是在微观级别(而不是在抽象概念的宏观级别，例如词类)上操作。在第二个阶段，我将通过将新方法应用于两组新数据来测试新方法的效率，这两组数据考察了在名词短语中具有不同功能的英语条目(如男孩、快乐鹦鹉等成分，用于识别实体)。这些集合涉及(1)项目，例如具有归因于属性(好看)的原始功能的项目，例如，其还具有新的功能，其中它强化了描述属性的另一项，例如，在‘相当糟糕的一周’中；以及(2)诸如‘Hard’的项目，其可以归因于属性(是坚实的，坚固的)，例如，‘Hard Rock’，但在‘硬奶酪’中开发了一种新功能，它指的是奶酪的一个亚类(如豪达等硬奶酪，而不是布里干酪等软奶酪)。我将展示微观分析方法改进了对单个项目和功能水平上的变化的描述。项目的第三阶段在新的微观分析描述的基础上重新构建了这些数据集中的项目可能发生的变化。其中一个主要问题是，新方法发现的机制如何与历史语言学中区分的两个主要变化机制有关：重新分析，即一种语言项目的句法被以不同的方式重新解释(例如，‘Hamburger’被分析为Hamburg-er，但被重新分析为Hamburg-Burger)，以及类比，即根据相似性将一种形式从一个环境扩展到另一个环境(例如，‘cheeseburger’和‘Fish burger’是在其重新分析后通过与‘Hamburger’类比而产生的)。我的假设是，类比在“漂亮”、“难”的变化中起着最重要的作用，但以不同的方式起作用，不同的元素充当类比的吸引者。

Linguists have always been interested in finding out how language changes. This area of research was transformed in the 1980s when electronic resources with large sets of historical data (corpora) became available for study. Corpora made it possible to study language change with quantitative methods for the first time. Historical corpus linguistics has focused on the development of more and better corpora and of quantitative methods. Methods for qualitative analysis have not been adapted to exploit all possibilities offered by the new resources. The main objective of this project is to redress this balance and put qualitative research and methodology on the agenda. It will do so by- proposing new qualitative methods to identify changes in historical corpus data, - showing the descriptive gain they offer, - showing that new data found with this methodology can bear on longstanding questions in historical linguistics such as the question how language change happens.First, I will develop new methods to identify one particular type of change, that by which linguistic items acquire a new function. An example is the change of the word 'certain', which originally only occurred in combinations such as 'a certain victory', where it means "sure, guaranteed" and its function is to describe a property of the victory in question. Through language change, 'certain' developed a new function as determiner: e.g. 'certain people don't like cats' where 'certain' is equivalent to 'some'. In this change, the form 'certain' remains the same. This raises the question how we can distinguish between the two functions in historical corpus data and which empirical evidence we can adduce for this change. The new methodology will consist of types of evidence relating to the form of the item and/or the range of contexts the item occurs in (its distribution) because changes to these aspects can be observed when comparing data for different periods. Because the methods investigate individual items, they are said to operate at the micro-level (rather than at the macro-level of abstract concepts such as e.g. word classes). In a second stage, I will test how efficient the new methodology is by applying it to two new sets of data which look at English items like 'certain' which have different functions in the noun phrase (a constituent such as 'a boy', 'the happy parrot', which identifies entities). These sets involve (1) items such as 'pretty' which has an original function of ascribing a property (being good-looking), e.g. 'a pretty girl', but which also has a new function in which it intensifies another item describing a property, e.g. in 'a pretty bad week' and (2) items such as 'hard', which can ascribe a property (being solid, firm), e.g. 'a hard rock', but has developed a new function in 'a hard cheese' where it indicates a subtype of cheese (hard cheeses such as gouda as opposed to soft cheeses such as brie). I will show that the micro-analytical methods improve the description of the changes on the level of the individual items and that of the functions.The third stage of the project reconstructs how change might have happened for the items in these data sets on the basis of the new micro-analytical descriptions. One of the main questions is how the mechanisms found with the new methodology relate to the two main mechanisms of change that have been distinguished in historical linguistics: reanalysis, i.e. the syntax of a linguistic item is reinterpreted in a different way (e.g. 'hamburger' was analysed as hamburg-er, but reanalysed as ham-burger), and analogy, i.e. the extension of a form from one environment to another based on similarity (e.g. 'cheeseburger' and 'fishburger' are created by analogy with 'hamburger' after its reanalysis). My hypothesis is that analogy plays the most important role in the changes of 'pretty', 'hard', but does so in different ways with different elements acting as attractor for the analogy.