Peter Uhrig

I don’t want to go all Yoko Ono on you

Abstract The present paper discusses what may be responsible for the creative effect that instances of a family of constructions, the go/get all/full [PROPER NOUN(S)] on Y pattern, appear to have on the hearer. For this purpose, a set of possible properties of creative language use will be discussed and checked against the results of a detailed corpus analysis of the pattern and related structures. Both the speaker’s and the hearer’s perspectives will be taken into account.

Why the Principle of No Synonymy is Overrated

Abstract The formulation of Goldberg’s oft-quoted Principle of No Synonymy is one of the factors responsible for a shift away in attention from alternations as postulated in the generative transformational tradition towards a view that regards the so-called alternatives as conveying different meanings and thus not being real alternatives. The rejection of the generativist position, in which one variant was regarded as primary and the other as derived from the primary variant, is of course justified and necessary in a cognitive linguistic approach, but it will be argued in this paper that the Principle of No Synonymy – if regarded as a dogma – is misleading in that it bears the risk of missing important generalisations across different patterns of the same verb. Furthermore, it will be argued that both linguistic variation and pre-emption are not perfectly compatible with the Principle of No Synonymy.

Chunk parsing in corpora

After a brief introduction to the concept of “chunk”, we describe the expectations in (“shallow”) chunk parsing from a syntactic and semantic perspective. For syntactic chunking, the task is separated in two steps, segmentation and prototyping. For segmentation, a sequence of words in a corpus is annotated by “IOB” tags (for inside, outside, and beginning of chunk), where the B tags are augmented by a POS (part of speech) tag. This has to be done either manually or by using an already annotated reference corpus in a semi-automatic fashion. An annotated corpus can be used for training a chunker by well-known techniques from pattern recognition. Setting up transformation-based learning as an iterative process can lead to precision and recall rates of about 94% on unseen corpora. For semantic chunking, additional information about the contents of phrases is required, e.g., by tagging noun phrases with person, location, etc. Experiences with tasks to identify thematic role fillers for verbs as agent, patient, or theme by shallow parsing are still significantly less successful than parsing with “full” syntax. 1 What on Earth is a Chunk? 1.1 Basic Features of Chunks A fundamental analytical task in Natural Language Processing (NLP) is the segmentation and labeling of texts. In a first step, texts are broken up into sentences as sequences of word forms (tokens). “Chunking” in general means to assign a partial structure to a sentence. “Tagging” assigns labels to the tokens which represent word specific and word form specific information such as the word category and morphological features. Chunk parsing regards sequences of tokens and tries to identify structural relations within and between segments. Chunk parsing as conceived by Abney [1] originated from a psycholinguistic motivation: “I begin with an intuition: when I read a sentence, I read it a chunk at a time. For example, the previous sentence breaks up something like this: (1) [I begin] [with an intuition]: [when I read] [a sentence], [I read it] [a chunk] [at a time] These chunks correspond in some way to prosodic patterns. It appears, for instance, that the strongest stresses in the sentence fall