The question of whether translated language constitutes a distinct linguistic variant has fueled debate within translation studies. The "third language" or "third code" concepts highlight the uniqueness of translated texts, distinct from both source and target languages. Baker's (1993) hypothesis of "translation universals" posits the existence of universal linguistic features in translated texts. While previous research primarily focused on lexical and grammatical features, this study addresses the gap in research on syntactic-semantic level universals. Explicitation, simplification, and leveling-out are translation universals that may be more prominent at the semantic level. This study aims to investigate these universals in English-Chinese translations using semantic role labeling (SRL) and textual entailment (TE) analysis to provide a more comprehensive understanding of the translation process and its impact on the syntactic-semantic structure of translated texts. The choice of SRL and TE methods is motivated by their ability to automate analysis without relying on deep semantic understanding and context, making them suitable for comparing large corpora of translated and non-translated texts. The study addresses two research questions: 1. Do translation universals exist at the syntactic-semantic level? If so, what are the typical syntactic-semantic features of translated texts? 2. What factors contribute to the distinct features observed in translated texts at the syntactic-semantic level?

The translation universal hypothesis, introduced by Baker (1993), has been subject to ongoing debate and refinement. Chesterman (2004) distinguished between T-universals (comparing translated texts with non-translated target language texts) and S-universals (comparing source and translated texts). Several sub-hypotheses have been proposed, including simplification, explicitation, normalization, and leveling-out. Explicitation, the tendency to explicate implicit information, is particularly relevant to semantic analysis. Previous studies often focused on lexical and grammatical levels, neglecting the potential for semantic-level explicitation. The lack of automated semantic analytical methods for large-scale corpora has hindered the exploration of translation universals from a syntactic-semantic perspective. This study overcomes this limitation by employing SRL and TE analysis, which avoid the content-specific biases of methods like word vector models.

This study employs a comparative corpus-based approach using three corpora: English source texts (ES), English-Chinese translations (CT), and non-translated Chinese original texts (CO). The Yiyan English-Chinese Parallel Corpus and the Lancaster Corpus of Mandarin Chinese were used for ES, CT, and CO respectively. Data was cleaned manually before analysis. Semantic role labeling was performed using N-LTP (for Chinese) and AllenNLP (for English). Six frequent semantic roles (A0, A1, A2, ADV, MNR, DIS) were selected for in-depth investigation. Textual entailment analysis, modified for this study, measured the distance between a given sentence (H) and its entailing sentence (T), created by substituting the predicate in H with its root hypernym. Wu-Palmer Similarity and Lin Similarity were used to quantify this distance. Five indices were used to represent syntactic-semantic features: average number of semantic roles per verb (ANPV), average number of semantic roles per sentence (ANPS), average role length (ARL), Wu-Palmer Similarity (Wup Sim), and Lin Similarity (Lin Sim). Statistical tests (Leneve's test and Mann-Whitney U test) were conducted to compare the indices across ES, CT, and CO. For syntactic subsumption analysis, standardized ARL was used to minimize the influence of differences in sentence length between English and Chinese. The study also examines the distributions of Wup Sim and Lin Sim to reveal nuanced differences beyond average values.

The results reveal significant differences in syntactic-semantic features between CT, ES, and CO. Concerning S-universals (comparing ES and CT), Mann-Whitney U tests showed significant differences across most indices. Regarding semantic subsumption, verbs in CT had lower Wup Sim and Lin Sim values than in ES, indicating a higher level of explicitation. This difference was partly explained by the increased use of notional verbs in CT replacing “be” verbs in ES and by the denominalization process. The analysis of syntactic subsumption showed that CT had lower ANPV and ARL values than ES, signifying simplification in argument structure. However, ANPS showed no significant difference. The analysis of specific semantic roles showed that agents (A0) and discourse markers (DIS) in CT were more frequent than in ES, confirming explicitation, while patients (A1) and recipients (A2) showed simplification. The examination of adverbials (ADV) and manners (MNR) further revealed a denominalizing effect in translation. Analyzing T-universals (comparing CT and CO), the study found subtle but statistically significant differences, particularly in the distributions of Wup Sim and Lin Sim. CT exhibited a more centralized distribution compared to CO, indicating leveling-out. Syntactic subsumption showed that CT had higher ANPV, ANPS, and ARL than CO, suggesting more complex and condensed syntactic-semantic structures in CT. Analysis of specific semantic roles in T-universals demonstrated that most of the syntactic-semantic features of CT are significantly higher than those in CO. The explicitation observed in agents (A0) and discourse markers (DIS) was consistent in both S-universals and T-universals. Overall, CT demonstrated an 'eclectic' characteristic, showing both explicitation and simplification, and suggesting a balance between source and target language influences.

The findings support the "third language" hypothesis and partially validate the translation universal hypothesis at the syntactic-semantic level. Explicitation, simplification, and leveling-out were observed, indicating that translation universals extend beyond lexical and grammatical features. The eclectic nature of CT's syntactic-semantic features, with contradictory results for S-universals and T-universals, points to the influence of both source and target languages. This supports the Hypothesis of Gravitational Pull, which posits that translated language is influenced by the target language's "magnetism effect," the source language's "gravitational pull effect," and a "connectivity effect." The consistent explicitation in agents (A0) and discourse markers (DIS) across S- and T-universals, however, suggests the influence of socio-cultural factors or translator-specific choices.

This study provides evidence for translation universals at the syntactic-semantic level, including explicitation, simplification, and leveling-out. The findings highlight the unique nature of translated texts, shaped by the interplay of source and target language influences, as well as socio-cultural factors. Future research should investigate other language pairs and refine the semantic analysis methods to further understand this complex interplay.

The study's findings are limited to English-Chinese translations. Future research should examine other language pairs to ascertain the generalizability of the results, particularly those with differing information structures. The analysis did not explore the interaction between semantic roles within argument structures, a factor that may affect semantic variations. Further studies should refine the semantic analysis tools used, improving the accuracy and depth of the analysis.