Current instruction-tuned language models are exclusively trained with textual preference data and thus may not be aligned to the unique requirements of other modalities, such as speech. To better align language models with the speech domain, we explore i) prompting strategies based on radio-industry best practices and ii) preference learning using a novel speech-based preference data of 20K samples collected by annotators who listen to response pairs. Both human and automatic evaluation show that both prompting and preference learning increase the speech-suitability of popular instruction tuned LLMs. More interestingly, we show that these methods are additive; combining them achieves the best win rates in head-to-head comparison, resulting in responses that are preferred or tied to the base model in 76.2% of comparisons on average. Lastly, we share lexical, syntactical, and qualitative analyses that elicit how our studied methods differ with baselines in generating more speech-suitable responses.

In this paper, we investigate which questions are challenging for retrieval-based Question Answering (QA). We (i) propose retrieval complexity (RC), a novel metric conditioned on the completeness of retrieved documents, which measures the difficulty of answering questions, and (ii) propose an unsupervised pipeline to measure RC given an arbitrary retrieval system.Our proposed pipeline measures RC more accurately than alternative estimators, including LLMs, on six challenging QA benchmarks. Further investigation reveals that RC scores strongly correlate with both QA performance and expert judgment across five of the six studied benchmarks, indicating that RC is an effective measure of question difficulty.Subsequent categorization of high-RC questions shows that they span a broad set of question shapes, including multi-hop, compositional, and temporal QA, indicating that RC scores can categorize a new subset of complex questions. Our system can also have a major impact on retrieval-based systems by helping to identify more challenging questions on existing datasets.

The efficacy of neural “retrieve and generate” systems is well established for question answering (QA) over unstructured text. Recent efforts seek to extend this approach to knowledge graph (KG) QA by converting structured triples to unstructured text. However, the relevance of KG triples retrieved by these systems limits their accuracy. In this paper, we improve the relevance of retrieved triples using a carefully designed re-ranker. Specifically, our pipeline (i) retrieves over documents of triples grouped by entity, (ii) re-ranks triples from these documents with context: triples in the 1-hop neighborhood of the documents’ subject entity, and (iii) generates an answer from highly relevant re-ranked triples. To train our re-ranker, we propose a novel “triple-level” labeling strategy that infers fine-grained labels and shows that these significantly improve the relevance of retrieved information. We show that the resulting “retrieve, re-rank, and generate” pipeline significantly improves upon prior KGQA systems, achieving a new state-of-the-art on FreebaseQA by 5.56% Exact Match. We perform multiple ablations that reveal the distinct benefits of our contextual re-ranker and labeling strategy and conclude with a case study that highlights opportunities for future works.