Speech Relation Extraction (SpeechRE) aims to extract relation triplets from speech data. However, existing studies usually use synthetic speech to train and evaluate SpeechRE models, hindering the further development of SpeechRE due to the disparity between synthetic and real speech. Meanwhile, the modality gap issue, unexplored in SpeechRE, limits the performance of existing models. In this paper, we construct two real SpeechRE datasets to facilitate subsequent researches and propose a Multi-level Cross-modal Alignment Model (MCAM) for SpeechRE. Our model consists of three components: 1) a speech encoder, extracting speech features from the input speech; 2) an alignment adapter, mapping these speech features into a suitable semantic space for the text decoder; and 3) a text decoder, autoregressively generating relation triplets based on the speech features. During training, we first additionally introduce a text encoder to serve as a semantic bridge between the speech encoder and the text decoder, and then train the alignment adapter to align the output features of speech and text encoders at multiple levels. In this way, we can effectively train the alignment adapter to bridge the modality gap between the speech encoder and the text decoder. Experimental results and in-depth analysis on our datasets strongly demonstrate the efficacy of our method.

Due to the continuous emergence of new data, version updates have become an indispensable requirement for Large Language Models (LLMs). The training paradigms for version updates of LLMs include pre-training from scratch (PTFS) and continual pre-training (CPT). Preliminary experiments demonstrate that PTFS achieves better pre-training performance, while CPT has lower training cost. Moreover, their performance and training cost gaps widen progressively with version updates. To investigate the underlying reasons for this phenomenon, we analyze the effect of learning rate adjustments during the two stages of CPT: preparing an initialization checkpoint and continual pre-training based on this checkpoint. We find that a large learning rate in the first stage and a complete learning rate decay process in the second stage are crucial for version updates of LLMs. Hence, we propose a learning rate path switching training paradigm. Our paradigm comprises one main path, where we pre-train a LLM with the maximal learning rate, and multiple branching paths, each of which corresponds to an update of the LLM with newly-added training data. Extensive experiments demonstrate the effectiveness and generalization of our paradigm. Particularly, when training four versions of LLMs, our paradigm reduces the total training cost to 58% compared to PTFS, while maintaining comparable pre-training performance.

To achieve non-parametric NMT domain adaptation, k-Nearest-Neighbor Machine Translation (kNN-MT) constructs an external datastore to store domain-specific translation knowledge, which derives a kNN distribution to interpolate the prediction distribution of the NMT model via a linear interpolation coefficient 𝜆. Despite its success, kNN retrieval at each timestep leads to substantial time overhead. To address this issue, dominant studies resort to kNN-MT with adaptive retrieval (kNN-MT-AR), which dynamically estimates 𝜆 and skips kNN retrieval if 𝜆 is less than a fixed threshold. Unfortunately, kNN-MT-AR does not yield satisfactory results. In this paper, we first conduct a preliminary study to reveal two key limitations of kNN-MT-AR: 1) the optimization gap leads to inaccurate estimation of 𝜆 for determining kNN retrieval skipping, and 2) using a fixed threshold fails to accommodate the dynamic demands for kNN retrieval at different timesteps. To mitigate these limitations, we then propose kNN-MT with dynamic retrieval (kNN-MT-DR) that significantly extends vanilla kNN-MT in two aspects. Firstly, we equip kNN-MT with a MLP-based classifier for determining whether to skip kNN retrieval at each timestep. Particularly, we explore several carefully-designed scalar features to fully exert the potential of the classifier. Secondly, we propose a timestep-aware threshold adjustment method to dynamically generate the threshold, which further improves the efficiency of our model. Experimental results on the widely-used datasets demonstrate the effectiveness and generality of our model.