Ambiguity may occur as homonymy, a word having two distinct meanings, as polysemy, a word having two related meanings, as vagueness, or structural ambiguity
...[Jäppinen et al., 1983,Jäppinen and Ylilammi, 1986]
Another practically complete model of Finnish morphology, the two-level model was developed by [Koskenniemi, 1983] in University of Helsinki. The two-level model is generally applicable over various languages and language families. Thus, prototypes of the two-level model have been implemented for over 30 languages. The most comprehensive implementations exist for Finnish, English, Swedish, Russian, Swahili, French, Arabic and Basque [Lindén, 1993]. A language-independent formalism, Constraint Grammar (CG) has also been developed for syntactic analysis [Karlsson, 1990,Karlsson et al., 1995c,Karlsson et al., 1995a,Karlsson et al., 1995b]. The recognition rate for a large English corpus, when parsing new unrestricted running text and after a morphological analysis by the two-level model, is approximately 98%, i.e., only 2 words out of 100 get the wrong syntactic code [Järvinen, 1994].
The term 'processing' is here used to refer to applications, whereas the term 'interpretation' emphasizes the cognitive point of view.
Collected by Ben Chi ($http://uacsc2.albany.edu{/\sim}bec/color.html$).
A definition of semantics and pragmatics that would be widely accepted is somewhat difficult to give. Levinson (1983) gives multiple possible definitions. One of the definitions is as follows: ``Pragmatics is the study of the ability of language users to pair sentences with the contexts in which they would appear.'' He further states that this definition fits well with the definition of semantics according to which semantic theories are concerned with the recursive assignment of truth conditions to well-formed expressions of language. One general aspect of defining semantics is that it specifies the relation between linguistic expressions and the referents of the expressions.
Radically connectionist means here an approach in which artificial neural networks are used so that they are adaptive, and the intermediate representations are numerical and their interpretation can only be based on the adaptation process.
In this work, words are handled as the original word forms appearing in the text
Linell (1982) has written about the written language bias in the following way: ``Our conception of linguistic behavior is biased by a tendency to treat processes, activities, and conditions of them in terms of object-like, static, autonomous and permanent structures, i.e., as if they shared such properties with written characters, words, texts, pictures and images. [...] In general, most of Western philosophy and science has been stuck with the metaphysical assumption that the world is made up of 'things' or 'objects'.''
In philosophy this view of the relationship between language and world was, among others, strongly proposed in the early works by Ludwig Wittgenstein. Perhaps his arrogance in Tractatus Logico-Philosophicus (saying that all the main problems are solved) was just premature.
A remark on the notion of 'symbol' may be necessary: the basic idea is to consider the possibility of grounding the symbols based on the unsupervised learning scheme. The symbols are used on the level of communication and may be used as the labels for the (usually) continuous multi-dimensional conceptual spaces. What is the role of these symbols in further processing is left open in this work. However, the requirement of grounding the symbols is considered to be crucial.
Symbol grounding, embodiment and their connectionist modeling is a central topic, e.g., in [Varela et al., 1993,Regier, 1995].
A connection to the relation between thought and language may be considered: if the weight of the linguistic input is high enough, the overall organization is partly determined by it rather than being only based on the ``perceptual'' component.
Not the present author.
The bibliography of the Self-Organizing Map (SOM) and Learning Vector Quantization (LVQ) compiled in the Neural Networks Research Centre at Helsinki University of Technology is available at the WWW address http://www.cis.hut.fi/nnrc/refs/.
The two standard measures of retrieval effectiveness are precision, the number of relevant retrieved documents over the total number of retrieved documents, and recall, the ratio of relevant retrieved documents to the total number of (known) relevant documents.
N-gram is a sequence of n characters.
A demonstration of the ET-Map is available at the WWW address
The effect of this smoothing has, however, been found to be rather small at least in some applications [Kaski, 1997a]
HyperText Markup Language
The demonstration can be seen in the WWW address http://websom.hut.fi/websom/.
Timo Honkela