Mitigating positional bias of language models (LMs) for listwise inputs is a well-known and important problem (e.g., lost-in-the-middle). While zero-shot order-invariant LMs have been proposed to solve this issue, their success on practical listwise problems has been limited. In this work, as a first contribution, we identify and overcome two limitations to make zero-shot invariant LMs more practical: (1) training and inference distribution mismatch arising from modifying positional ID assignments to enforce invariance, and (2) failure to adapt to a mixture of order-invariant and sensitive inputs in practical listwise problems. To overcome, we propose (1) RoToR, a zero-shot invariant LM for genuinely order-invariant inputs with minimal modifications of positional IDs, and (2) Selective Routing, an adaptive framework that handles both order-invariant and order-sensitive inputs in listwise tasks. On the Lost in the middle (LitM), Knowledge Graph Question Answering (KGQA), and MMLU benchmarks, we show that RoToR with Selective Routing can effectively handle practical listwise input tasks in a zero-shot manner.

Multimodal abstractive summarization with sentence output is to generate a textual summary given a multimodal triad -- sentence, image and audio, which has been proven to improve users satisfaction and convenient our life. Existing approaches mainly focus on the enhancement of multimodal fusion, while ignoring the unalignment among multiple inputs and the emphasis of different segments in feature, which has resulted in the superfluity of multimodal interaction. To alleviate these problems, we propose a Multimodal Hierarchical Selective Transformer (mhsf) model that considers reciprocal relationships among modalities (by low-level cross-modal interaction module) and respective characteristics within single fusion feature (by high-level selective routing module). In details, it firstly aligns the inputs from different sources and then adopts a divide and conquer strategy to highlight or de-emphasize multimodal fusion representation, which can be seen as a sparsely feed-forward model - different groups of parameters will be activated facing different segments in feature. We evaluate the generalism of proposed mhsf model with the pre-trained+fine-tuning and fresh training strategies. And Further experimental results on MSMO demonstrate that our model outperforms SOTA baselines in terms of ROUGE, relevance scores and human evaluation.