An Account for Compound Prepositions in Farsi

Zahra Abolhassani ChimeACL 2006

TL;DR

This paper tackles the linguistic ambiguity of "Preposition + Noun" constructions in Farsi, which often behave like single compound units rather than phrases. By applying the theory of incorporation, the author provides a formal syntactic framework to distinguish these compounds, offering a clearer path for more accurate part-of-speech (POS) tagging and parsing in Natural Language Processing (NLP).

What Problem Does This Paper Address?

In many languages, prepositions are considered a "closed set"—a fixed list of words like in, on, or at. However, Farsi features a productive class of constructions where a simple preposition combines with a noun (e.g., bar asās-e, meaning "on the basis of") to function as a single prepositional unit.

The problem is that these constructions are structurally identical to standard Prepositional Phrases (PPs). For a computer trying to parse Farsi, this creates a headache: Is the word "basis" a noun that can be modified by adjectives and plurals, or is it a frozen part of a compound preposition? Traditional Farsi linguistics has been divided on this, and the lack of a formal account leads to errors in automated tagging and syntactic analysis.

Methods

The author employs a two-step approach:

  1. Structural/Constituency Testing: The paper uses a battery of linguistic tests to prove that the nouns in these specific constructions have lost their "noun-hood." These tests include checking for:
  2. Premodifiers: Can you add a demonstrative (this) or a superlative (newest) before the noun?
  3. Post-modifiers: Can you add an adjective or a plural marker to the noun?
  4. Syntactic Movement: Can the noun be "topicalized" (moved to the front of the sentence) or coordinated with another noun?
  5. Generative Framework (Incorporation): Drawing on Mark Baker’s (1988) theory of incorporation, the author proposes that these compounds are formed when a Noun head ($N^0$) moves and attaches itself to a Preposition head ($P^0$). This creates a "syntactic compound" where the noun is "sucked into" the preposition, changing how the word governs the rest of the sentence.

Datasets

Not discussed in the paper. The research is theoretical and qualitative, relying on constructed linguistic examples and established grammatical rules rather than training on a specific digital corpus or dataset.

Key Insights and Ideas

  • The "Non-Referential" Noun: The most critical insight is that for incorporation to happen, the noun must be abstract and "non-referential." In the compound "on the basis of," the word "basis" doesn't refer to a physical object; it has been bleached of its independent meaning to serve a functional role.
  • Phonology is a Dead End: The author notes that in Farsi, accent patterns for these compounds are identical to regular phrases. Therefore, we cannot rely on how a word sounds to determine if it's a compound; we must look at its syntactic behavior.
  • Transition Phase: The paper suggests these structures are in a "transition phase" from phrases to fixed morphological compounds. They aren't quite single words yet, but they no longer behave like multi-word phrases.
  • The "Head Escape Principle": The paper highlights that while the head of a phrase usually allows the accent to fall on the farthest member, these compounds act as a single head, which simplifies the "governance" of the sentence in a computational model.

Key Results

The paper concludes that "Preposition + Noun" combinations in Farsi should be treated as Compound Prepositions (CPs) if they meet specific criteria:

  1. The preposition must be "true" and morphologically simple (e.g., dar, be, az).
  2. The noun must be abstract, simple, and incapable of taking modifiers.

The author provides an "algorithm-like" flowchart (Figure 3) for a Lexicon Checker. This tool would take a Noun and Preposition as input, check them against a list of semantic and morphological features (like [+abstract], [-referential]), and determine whether the NLP system should label the pair as a Compound Preposition or a standard Prepositional Phrase.

Limitations and Future Work

  • Lack of Quantitative Validation: While the theoretical framework is robust, the paper does not provide experimental results showing how much this model improves the accuracy of an actual Farsi parser or POS tagger.
  • Abstract Incorporation: The author admits that incorporation in Farsi is "abstract," meaning there is no overt movement of words that a reader can see; it is a structural change that must be inferred.
  • Future Directions: The author suggests that this model should be integrated into Natural Language Processing applications to assist in word-sense disambiguation and information retrieval (IR), as knowing the correct POS is vital for stemming and keyword extraction.

Summary updated

BibTeX
@inproceedings{chime2006account,
  author = {Zahra Abolhassani Chime},
  title = {{An Account for Compound Prepositions in Farsi}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2006},
}
Figure