International Journal of Computational Linguistics & Chinese Language Processing
Vol. 2, No. 2, August 1997


Title:
Building a Bracketed Corpus Using X2 Statistics

Author:
Yue-Shi Lee, Hsin-Hsi Chen

Abstract:
Research based on treebanks is ongoing for many natural language applications. However, the work involved in building a large-scale treebank is laborious and time-consuming. Thus, speeding up the process of building a treebank has become an important task. This paper proposes two versions of probabilistic chunkers to aid the development of a bracketed corpus. The basic version partitions part-of-speech sequences into chunk sequences, which form a partially bracketed corpus. Applying the chunking action recursively, the recursive version generates a fully bracketed corpus. Rather than using a treebank as a training corpus, a corpus, which is tagged with part-of-speech information only, is used. The experimental results show that the probabilistic chunker has a correct rate of more than 94% in producing a partially bracketed corpus and also gives very encouraging results in generating a fully bracketed corpus. These two versions of chunkers are simple but effective and can also be applied to many natural language applications.

Keyword:
Bracketed Corpus, Probabilistic Chunkers, Treebank, X2 Statistics


Title:
Longest Tokenization

Author:
Jin Guo

Abstract:
Sentence tokenization is the process of mapping sentences from character strings into strings of tokens. This paper sets out to study longest tokenization which is a rich family of tokenization strategies following the general principle of maximum tokenization. The objectives are to enhance the knowledge and understanding of the principle of maximum tokenization in general, and to establish the notion of longest tokenization in particular. The main results are as follows: (1) Longest tokenization, which takes a token n-gram as a tokenization object and seeks to maximize the object length in characters, is a natural generalization of the Chen and Liu Heuristic on the table of maximum tokenizations. (2) Longest tokenization is a rich family of distinct and unique tokenization strategies with many widely used maximum tokenization strategies, such as forward maximum tokenization, backward maximum tokenization, forward-backward maximum tokenization, and shortest tokenization, as its members. (3) Longest tokenization is theoretically a true subclass of critical tokenization, as the essence of maximum tokenization is fully captured by the latter. (4) Longest tokenization is practically the same as shortest tokenization, as the essence of length-oriented maximum tokenization is captured by the latter. Results are obtained using both mathematical examination and corpus investigation.

Keyword:
sentence tokenization, tokenization disambiguation, maximum tokenization, critical tokenization, word segmentation, word identification


Title:
Segmentation Standard for Chinese Natural Language Processing

Author:
Chu-Ren Huang, Keh-jiann Chen, Feng-yi Chen, and Li-Li Chang

Abstract:
This paper proposes a segmentation standard for Chinese natural language processing. The standard is proposed to achieve linguistic felicity, computational feasibility, and data uniformity. Linguistic felicity is maintained by a definition of segmentation unit that is equivalent to the theoretical definition of word, as well as a set of segmentation principles that are equivalent to a functional definition of a word. Computational feasibility is ensured by the fact that the above functional definitions are procedural in nature and can be converted to segmentation algorithms as well as by the implementable heuristic guidelines which deal with specific linguistic categories. Data uniformity is achieved by stratification of the standard itself and by defining a standard lexicon as part of the standard.


Title:
Aligning More Words with High Precision for Small Bilingual Corpora

Author:
Sue J. Ker, Jason S. Chang

Abstract:
In this paper, we propose an algorithm for identifying each word with its translations in a sentence and translation pair.

Previously proposed methods require enormous amounts of bilingual data to train statistical word-by-word translation models. By taking a word-based approach, these methods align frequent words with consistent translations at a high precision rate. However, less frequent words or words with diverse translations generally do not have statistically significant evidence for confident alignment. Consequently, incomplete or incorrect alignments occur. Here, we attempt to improve on the coverage using class-based rules. An automatic procedure for acquiring such rules is also described. Experimental results confirm that the algorithm can align over 85% of word pairs while maintaining a comparably high precision rate, even when a small corpus is used in training.

Keyword:
Word alignment, machine readable dictionary and thesaurus, bilingual corpus, word sense disambiguation


Title:
An Unsupervised Iterative Method for Chinese New Lexicon Extraction

Author:
Jing-Shin Chang, Keh-Yih Su

Abstract:
An unsupervised iterative approach for extracting a new lexicon (or unknown words) from a Chinese text corpus is proposed in this paper. Instead of using a non-iterative segmentation-merging-filtering-and-disambiguation approach, the proposed method iteratively integrates the contextual constraints (among word candidates) and a joint character association metric to progressively improve the segmentation results of the input corpus (and thus the new word list.) An augmented dictionary, which includes potential unknown words (in addition to known words), is used to segment the input corpus, unlike traditional approaches which use only known words for segmentation. In the segmentation process, the augmented dictionary is used to impose contextual constraints over known words and potential unknown words within input sentences; an unsupervised Viterbi Training process is then applied to ensure that the selected potential unknown words (and known words) maximize the likelihood of the input corpus. On the other hand, the joint character association metric (which reflects the global character association characteristics across the corpus) is derived by integrating several commonly used word association metrics, such as mutual information and entropy, with a joint Gaussian mixture density function; such integration allows the filter to use multiple features simultaneously to evaluate character association, unlike traditional filters which apply multiple features independently. The proposed method then allows the contextual constraints and the joint character association metric to enhance each other; this is achieved by iteratively applying the joint association metric to truncate unlikely unknown words in the augmented dictionary and using the segmentation result to improve the estimation of the joint association metric. The refined augmented dictionary and improved estimation are then used in the next iteration to acquire better segmentation and carry out more reliable filtering. Experiments show that both the precision and recall rates are improved almost monotonically, in contrast to non-iterative segmentation-merging-filtering-and-disambiguation approaches, which often sacrifice precision for recall or vice versa.

With a corpus of 311,591 sentences, the performance is 76% (bigram), 54% (trigram), and 70% (quadragram) in F-measure, which is significantly better than using the non-iterative approach with F-measures of 74% (bigram), 46% (trigram), and 58% (quadragram).

Keyword:
Unknown Word Identification, New Lexicon Extraction, Unsupervised Method, Iterative Enhancement, Chinese, Lexicon