Pointer-decoded Seq2Seq transformers remain a strong recipe for aspect-based sentiment analysis (ABSA), but native German Seq2Seq backbones remain scarce, and current pretraining trends favor decoder-only architectures. Existing assembly recipes disagree on which pairings work. We therefore study models assembled from independently pretrained encoder and decoder LMs, primarily on German ABSA, where fixed pointer outputs expose both tuple accuracy and grammar violations. All nine backbone combinations train under both separate and parameter-sharing merged cross-attention, with separate attention slightly stronger overall. Depth and scale sweeps show that additional capacity pays off more on the encoder side for these pointer-decoded tasks, and our strongest two-stack assembly trails mBART but outperforms mT5-small on both German datasets. Motivated by decoder-only pretraining, we replace the encoder and decoder with one backbone that encodes the source and autoregressively produces pointer queries. Bidirectional source visibility and scale matter more consistently than pointer-query refinement: a 120M LLäMmlein model exceeds both same-family two-stack means with roughly half the parameters, and a 1B ModernGBERT reaches mBART-level performance.

We revisit the BARTABSA framework for aspect-based sentiment analysis with modern decoder LLMs to assess the importance of explicit structure modeling today. Our updated implementation - BARTABSA++ - features architectural enhancements that boost performance and training stability.Systematic testing with various encoder-decoder architectures shows that BARTABSA++ with BART-Large achieves state-of-the-art results, even surpassing a finetuned GPT-4o model.Our analysis indicates the encoder’s representational quality is vital, while the decoder’s role is minimal, explaining the limited benefits of scaling decoder-only LLMs for this task. These findings underscore the complementary roles of explicit structured modeling and large language models, indicating structured approaches remain competitive for tasks requiring precise relational information extraction.

Entity-aware machine translation faces significant challenges when translating culturally-adapted named entities that require knowledge beyond the source text. We present SALT (SQL-based Approach for LLM-Free Entity-Aware-Translation), a parameter-efficient system for the SemEval-2025 Task 2. Our approach combines SQL-based entity retrieval with constrained neural translation via logit biasing and explicit entity annotations. Despite its simplicity, it achieves state-of-the-art performance (First Place) among approaches not using gold-standard data, while requiring far less computation than LLM-based methods. Our ablation studies show simple SQL-based retrieval rivals complex neural models, and strategic model refinement outperforms increased model complexity. SALT offers an alternative to resource-intensive LLM-based approaches, achieving comparable results with only a fraction of the parameters.