Abstract
The information conveyed by some sentences would be more easily understood by a reader if it were expressed in multiple sentences. We call such sentences content heavy: these are possibly grammatical but difficult to comprehend, cumbersome sentences. In this paper we introduce the task of detecting content-heavy sentences in cross-lingual context. Specifically we develop methods to identify sentences in Chinese for which English speakers would prefer translations consisting of more than one sentence. We base our analysis and definitions on evidence from multiple human translations and reader preferences on flow and understandability. We show that machine translation quality when translating content heavy sentences is markedly worse than overall quality and that this type of sentence are fairly common in Chinese news. We demonstrate that sentence length and punctuation usage in Chinese are not sufficient clues for accurately detecting heavy sentences and present a richer classification model that accurately identifies these sentences.
Introduction
In Natural Language Processing, "content-heavy" sentences are defined as those that are technically grammatical but so information-dense that they impede comprehension. This phenomenon is particularly prevalent in Chinese-to-English translation. Chinese syntax frequently employs commas where English conventions would demand a full stop, resulting in long, "run-on" structures. When machine translation (MT) systems attempt to preserve these complex structures in a single English sentence, the output is often disfluent and difficult to parse.
Defining "Heaviness" through Human Consensus
To objectively identify content-heavy sentences, the researchers utilized a consensus-based approach. A Chinese sentence was labeled "heavy" if at least two out of four professional human translators chose to split it into multiple English sentences. This metric reflects a collective human judgment that the source sentence contains too much information for a single English sentence to handle gracefully.
As shown in the data distribution, a significant portion of sentences in standard benchmarks (approximately 27.2%) meet this criteria for heaviness, indicating that this is not a marginal issue but a common characteristic of Chinese news text.
The Machine Translation Performance Gap
The study confirms that content-heaviness is a major bottleneck for current MT technology. There is a stark performance degradation when translating heavy sentences compared to non-heavy ones.
The analysis revealed a 3.9 BLEU point difference between these categories. This gap suggests that MT systems struggle to maintain coherence and fluency when forced to map dense Chinese structures into single English sentences.
Detection Methodology
The researchers developed a binary classification model using Logistic Regression to identify these sentences automatically. Key features included:
- Syntactic Weight: Analyzing dependency structures, specifically the "weight" of modifiers attached to nouns.
- Comma Disambiguation: Features designed to distinguish between commas acting as simple pauses and those functioning as sentence boundaries.
- Structural Cues: Tracking punctuation, conjunctions, and specific particles (e.g., the Chinese "DE").
While sentence length is often assumed to be the primary driver of complexity, the model demonstrated that length alone is an insufficient predictor. The full model achieved 80.73% accuracy, significantly outperforming a length-only baseline (71.43%).
Conclusion and Future Work
This research establishes "content-heaviness" as a distinct linguistic property that can be modeled and detected with high precision. By identifying these "linguistic traffic jams," we can improve translation pipelines. Future work will focus on developing automated splitting algorithms to simplify source text before translation, ensuring that the final output respects the cognitive load of the reader.
Summary updated