scispace - formally typeset
Search or ask a question
Topic

Chunking (computing)

About: Chunking (computing) is a research topic. Over the lifetime, 577 publications have been published within this topic receiving 11036 citations.


Papers
More filters
Book ChapterDOI
01 Jan 1999
TL;DR: This work has shown that the transformation-based learning approach can be applied at a higher level of textual interpretation for locating chunks in the tagged text, including non-recursive “baseNP” chunks.
Abstract: Transformation-based learning, a technique introduced by Eric Brill (1993b), has been shown to do part-of-speech tagging with fairly high accuracy. This same method can be applied at a higher level of textual interpretation for locating chunks in the tagged text, including non-recursive “baseNP” chunks. For this purpose, it is convenient to view chunking as a tagging problem by encoding the chunk structure in new tags attached to each word. In automatic tests using Treebank-derived data, this technique achieved recall and precision rates of roughly 93% for baseNP chunks (trained on 950K words) and 88% for somewhat more complex chunks that partition the sentence (trained on 200K words). Working in this new application and with larger template and training sets has also required some interesting adaptations to the transformation-based learning approach.

1,236 citations

Proceedings ArticleDOI
13 Sep 2000
TL;DR: The CoNLL-2000 shared task: dividing text into syntactically related non-overlapping groups of words, so-called text chunking is described.
Abstract: We describe the CoNLL-2000 shared task: dividing text into syntactically related non-overlapping groups of words, so-called text chunking. We give background information on the data sets, present a general overview of the systems that have taken part in the shared task and briefly discuss their performance.

855 citations

Proceedings ArticleDOI
31 Aug 2002
TL;DR: A new statistical Japanese dependency parser using a cascaded chunking model that is simple and efficient, since it parses a sentence deterministically only deciding whether the current segment modifies the segment on its immediate right hand side.
Abstract: In this paper, we propose a new statistical Japanese dependency parser using a cascaded chunking model. Conventional Japanese statistical dependency parsers are mainly based on a probabilistic model, which is not always efficient or scalable. We propose a new method that is simple and efficient, since it parses a sentence deterministically only deciding whether the current segment modifies the segment on its immediate right hand side. Experiments using the Kyoto University Corpus show that the method outperforms previous systems as well as improves the parsing and training efficiency.

524 citations

Book ChapterDOI
11 Jan 2008
TL;DR: There are normative descriptions of stages of L2 proficiency that were drawn up in as atheoretical way as possible by the American Council on the Teaching of Foreign Languages (ACTFL) (Higgs, 1984) and there are several relevant case studies of child SLA.
Abstract: schema. For a general summary, there are normative descriptions of stages of L2 proficiency that were drawn up in as atheoretical way as possible by the American Council on the Teaching of Foreign Languages (ACTFL) (Higgs, 1984). These Oral Proficiency Guidelines include the following descriptions of novice and intermediate levels that emphasize the contributions of patterns and formulae to the development of later creativity: “Novice Low: Oral production consists of isolated words and perhaps a few highfrequency phrases... Novice High: Able to satisfy partially the requirements of basic communicative exchanges by relying heavily on learned utterances but occasionally expanding these through simple recombinations of their elements... Intermediate: The intermediate level is characterized by an ability to create with the language by combining and recombining learned elements, though primarily in a reactive mode “ (ACTFL, 1986). Constructions, chunking, and connectionism p. 19 Thus the ACTFL repeatedly stresses the constructive potential of collocations and chunks of language. This is impressive because the ACTFL guidelines were simply trying to describe SLA as objectively as possible—there was no initial theoretical focus on formulae -yet nonetheless the role of formulae became readily apparent in the acquisition process. There are several relevant case studies of child SLA. Wong-Fillmore (1976) presented the first extensive longitudinal study that focused on formulaic language in L2 acquisition. Her subject, Nora, acquired and overused a few formulaic expressions of a new structural type during one period, and then amassed a variety of similar forms during the next: previously unanalyzed chunks became the foundations for creative construction (see also Vihman’s, 1980 analyses of her young son Virve’s SLA). Such observations of the formulaic beginnings of child L2 acquisition closely parallel those of Pine and Lieven for L1. There are a few studies which focus on these processes in classroom-based SLA. R. Ellis (1984) described how three classroom learners acquired formulas which allowed them to meet their basic communicative needs in an ESL classroom, and how the particular formulas they acquired reflected input frequency -they were those which more often occurred in the social and organizational contexts that arose in the classroom environment. Weinert (1994) showed how English learners’ early production of complex target-like German FL negation patterns came through the memorization of complex forms in confined linguistic contexts, and that some of these forms were used as a basis for extension of patterns. Myles, Mitchell and Hooper (1998, 1999) describe the first two years of development of interrogatives in a classroom of anglophone French L2 Constructions, chunking, and connectionism p. 20 beginners, longitudinally tracking the breakdown of formulaic chunks such as comment t’appelles-tu?, comment s’appelle-t-il? and òu habites-tu?, the creative construction of new interrogatives by recombination of their parts, and the ways in which formulae fed the constructive process. Bolander (1989) analyzed the role of chunks in the acquisition of inversion in Swedish by Polish, Finnish and Spanish immigrants enrolled in a fourmonth intensive course in Swedish. In Swedish, the inversion of subject-verb after a sentence-initial non-subject is an obligatory rule. Bolander identified the majority of the inversion cases in her data as being of a chunk-like nature with a stereotyped reading such as det kan man säga (that can one say) and det tycker jag (so think I). Inversion in these sort of clauses is also frequent when the object is omitted as in kan man säga (one can say) and tycker jag (think I), and this pattern was also well integrated in the interlanguage of most of these learners. Bolander showed that the high accuracy on these stereotyped initial-object clauses generalized to produce a higher rate of correctness on clauses with non-stereotyped initial objects than was usual for other types of inversion clause in her data, and took this as evidence that creative language was developing out of familiar formulae. Although there are many reviews which discuss the important role of formula use in SLA (e.g., Hakuta 1974; Nattinger & DeCarrico, 1992; Towell & Hawkins, 1994; Weinert, 1995; and Wray, 1992), there is clearly further need for larger-sampled SLA corpora which will allow detailed analysis of acquisition sequences. De Cock (1998) presents analyses of corpora of language-learner productions using automatic recurrent sequence extractions. These show that second language learners use formulae at least as much as native speakers and at times at significantly higher rates. There is much promise Constructions, chunking, and connectionism p. 21 of such computer-based learner corpus studies (Granger, 1998), providing that sufficient trouble is taken to gather the necessarily intensive longitudinal learner data. There is also need to test the predictions of usage-based theories regarding the influences of typeand tokenfrequency as they apply in SLA. 3 Psychological Accounts of Associative Learning This section concerns the psychological learning mechanisms which underpin the acquisition of constructions. Constructivists believe that language is cut of the same cloth as other forms of learning: although it differs importantly from other knowledge in its specific content and problem space, it is acquired using generic learning mechanisms. The Law of Contiguity, the most basic principle of association, pervades all aspects of the mental representation of language: “Objects once experienced together tend to become associated in the imagination, so that when any one of them is thought of, the others are likely to be thought of also, in the same order of sequence or coexistence as before.” (James, 1890, p. 561). 3.1 Chunking What’s the next letter in a sentence beginning ‘T...’? You know it is much more likely to be h, or a vowel, than it is z or other consonants. You know it couldn’t be q. But I’ll warrant you have never been taught this. What is the first word in that sentence? You are likely to plump for the, or that, rather than thinks or theosophy. If ‘The...’, how does it continue? ‘With an adjective or noun’, you might reply. If ‘The cat...’, then what? And then again, complete ‘The cat sat on the...’. Fluent native speakers know a tremendous Constructions, chunking, and connectionism p. 22 amount about the sequences of language at all grains. We know how letters tend to cooccur (common bigrams, trigrams, and other orthographic regularities). We know the phonotactics of our tongue. We know phrase structure regularities. We know thousands of concrete collocations and we know abstract generalizations that derive from them. We have learned to chunk letters, sounds, morphemes, words, phrases, clauses, bits of cooccurring language at all levels. Psycholinguistic experiments show that we are tuned to these regularities in that we process faster and more easily language which accords with the expectations that have come from our unconscious analysis of the serial probabilities in our lifelong history of input (Ellis, in press). We learn these chunks from the very beginnings of learning a second language. Ellis, Lee and Reber (1999) observed people reading their first 64 sentences of a foreign language. While they read, they saw the referent of each sentence, a simple action sequence involving colored geometrical shapes, for example, the sentence ‘miu-ra ko-gi pye-ri lon-da’ was accompanied by a cartoon showing a square moving onto red circles. A linguistic description of this language might include the facts that it is an SOV language, it has adjective-noun word order, obligatory grammatical number (singular/plural) agreement in the form of matching suffix endings of a verb and its subject and of a noun and the adjective that modifies it, that the 64 sentences are all of the type: [N]Subject [A N]Object V, and that lexis was selected from a very small set of 8 words. But such explicit metalinguistic knowledge is not the stuff of early language acquisition. What did the learners make of it? To assess their intake, immediately after seeing each sentence, learners had to repeat as much as they could of it. How did their intake change over time? It gradually improved in all respects. With increasing exposure, Constructions, chunking, and connectionism p. 23 so performance incremented on diverse measures: the proportion of lexis correctly recalled, correct expression of the adjective-noun agreement, correct subject-verb agreement, totally correct sentence, number of correct bigrams and trigrams, and, generally, conformity to the sequential probabilities of the language at letter, word and phrase level. With other measures it was similarly apparent that there was steady acquisition of form-meaning links and of generalisable grammatical knowledge that allowed success on grammaticality judgement tests which were administered later (Ellis et al., 1999). To greater or lesser degree, these patterns, large and small, were being acquired simultaneously and collaboratively. Acquisition of these sequential patterns is amenable to explanation in terms of psychological theories of chunking. The notion of chunking has been at the core of shortterm memory research since Miller (1956) first proposed the term: while the chunk capacity of short-term memory (STM) is fairly constant at 7±2 chunks, its information capacity can be increased by chunking, a useful representational process in that low-level features that co-occur can be organized together and thence referred to as an individual entity. Chunking underlies your superior short-term memory for a patterned phone numbers (e.g. 0800-123777) or letter strings (e.g. AGREEMENTS, or FAMONUBITY) than for a more random sequences (e.g. 4957-632518, CXZDKLWQPM) even though all strings contain the sa

457 citations

Journal ArticleDOI
TL;DR: The hippocampal cortex has the capacity for chunking, but the hippocampal (limbic) arousal system plays a critical role in this chunking process by differentiall y priming (partially activating) free, as opposed to bound, neurons as discussed by the authors.
Abstract: Horizontal versus vertical associative memory concepts are denned. Vertical associative memory involves chunking: the specification of new (previously free) nodes to represent combinations of old (bound) nodes. Chunking is the basis of semantic memory, configuring in conditioning, and cognitive (as opposed to stimulus-response) learning. The cortex has the capacity for chunking, but the hippocampal (limbic) arousal system plays a critical role in this chunking process by differentiall y priming (partially activating) free, as opposed to bound, neurons. Binding a neuron produces negatively accelerated repression of its connections to the hippocampal arousal system, consolidating the memory by protecting the newly bound neuron from diffuse hippocampal input and thus retarding forgetting. Disruption of the hippocampal arousal system produces the amnesic syndrome of an inability to do new chunking (cognitive learning)—anterograde amnesia—and an inability to retrieve recently specified chunks—retrograde amnesia.

454 citations


Network Information
Related Topics (5)
Machine translation
22.1K papers, 574.4K citations
84% related
Parsing
21.5K papers, 545.4K citations
84% related
Language model
17.5K papers, 545K citations
82% related
Sentiment analysis
22.1K papers, 460.8K citations
81% related
Natural language
31.1K papers, 806.8K citations
80% related
Performance
Metrics
No. of papers in the topic in previous years
YearPapers
202125
202014
201931
201821
201735
201638