Automatic Translation With Attribute Grammars

Werner DilgerCOLING 1980

TL;DR

This paper presents a method for automatically generating attribute grammars from existing ATN grammars and translation rules. This approach aims to improve the efficiency and clarity of natural language translation systems by transforming them into a more structured and manageable attribute grammar framework.

What Problem Does This Paper Address?

The paper addresses the limitations of a specific natural language translation algorithm used in the PLIDIS system. This algorithm, while straightforward to implement, suffers from inefficiencies: it runs separately from parsing, requires TR-rules to explicitly handle parse tree structures, and makes it difficult to detect circularities. The authors propose using attribute grammars as a more robust and efficient alternative.

Methods

The core of the paper's method involves transforming an Augmented Transition Network (ATN) grammar, which is used for parsing, into a context-free grammar. This is achieved by deriving context-free productions from the ATN networks. Subsequently, the translation rules (TR-rules) are converted into attributes and semantic functions associated with these productions. The authors emphasize that this process is straightforward and can be illustrated with examples. They also note that by using attributes, the context-sensitivity of the original ATN networks can be preserved within the context-free framework.

Datasets

Not discussed in the paper. The paper uses example German questions related to water pollution control as illustrations.

Key Insights and Ideas

  • Bridging ATNs and Attribute Grammars: The paper demonstrates a practical method for converting ATN grammars, which are powerful for parsing but can be complex to manage for translation, into attribute grammars, which offer a more structured approach to semantic interpretation and translation.
  • Preserving Context-Sensitivity: A key insight is that the context-sensitivity inherent in ATN networks, often considered their main advantage, can be retained in the derived context-free productions by augmenting them with appropriate attributes and semantic functions.
  • Simplifying Translation Rule Management: By translating TR-rules into attributes and semantic functions, the system moves away from explicit tree-structure manipulation within rules, leading to a cleaner and potentially more efficient translation process.
  • Detecting Circularities: The attribute grammar framework offers a more formal basis for detecting circularities in the translation process, a problem that was difficult to address with the original TR-rule-based system.

Key Results

The paper provides a detailed walkthrough of the transformation process using example German questions. It illustrates how parse trees are generated from the transformed grammar and how translation rules are applied to derive predicate calculus expressions. While specific quantitative results (like speedup percentages) are not presented, the paper qualitatively demonstrates the feasibility and advantages of the attribute grammar approach for improving translation efficiency and manageability.

Limitations and Future Work

The authors acknowledge that the TR-rules, while powerful, can be cumbersome to write as they must account for all possible occurrences of a symbol. The attribute grammar approach, while more structured, requires careful definition of attributes and semantic functions. Future work could involve further improvements by simplifying productions, potentially by eliminating word-class symbols, and by developing well-defined algorithms for detecting translation circularities. The paper also implicitly suggests that this method can be applied to other natural language processing tasks beyond translation.

Summary updated

BibTeX
@inproceedings{dilger1980automatic,
  author = {Werner Dilger},
  title = {{Automatic Translation With Attribute Grammars}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1980},
}

Loading related content...

Figure