Reinforcement learning with natural language action space often suffers from the curse of dimensionality due to the combinatorial nature of the natural language. Previous research leverages pretrained language models to capture action semantics and reduce the size of the action space. However, since pretrained models are typically trained on general corpora, there can be an unpredictable mismatch between the priors encoded in pretrained models and the characteristics of the specific RL environment. To address this issue, we propose Mutual-Information Regularized Policy Optimization, MIPO. MIPO enables implicit and dynamic reduction of the action space. Starting from the prior provided by the pretrained language model, our method dynamically adjusts the prior during the learning process based on the guidance of mutual information regularization. Theoretically, we demonstrate that this policy optimization process leads to the monotonic improvement on the mutual-information regularized RL objective. Empirically, we conduct experiments in various environments and demonstrate the effectiveness of MIPO.

The effectiveness of large language models (LLMs) is often hindered by duplicated data in their extensive pre-training datasets. Current approaches primarily focus on detecting and removing duplicates, which risks the loss of valuable information and neglects the varying degrees of duplication. To address this, we propose a soft deduplication method that maintains dataset integrity while selectively reducing the sampling weight of data with high commonness. Central to our approach is the concept of “data commonness”, a metric we introduce to quantify the degree of duplication by measuring the occurrence probabilities of samples using an n-gram model. Empirical analysis shows that this method significantly improves training efficiency, achieving comparable perplexity scores with at least a 26% reduction in required training steps. Additionally, it enhances average few-shot downstream accuracy by 1.77% when trained for an equivalent duration. Importantly, this approach consistently improves performance, even on rigorously deduplicated datasets, indicating its potential to complement existing methods and become a standard pre-training process for LLMs.

This paper investigates the use of standard and non-standard adverbial markers in modern Chinese literature. In Chinese, adverbials can be derived from many adjectives, adverbs and verbs with the suffix “de”. The suffix has a standard and a non-standard written form, both of which are frequently used. Contrastive research on these two competing forms has mostly been qualitative or limited to small text samples. In this first large-scale quantitative study, we present statistics on 346 adverbial types from an 8-million-character text corpus drawn from Chinese literature in the 20th century. We present a semantic analysis of the verbs modified by adverbs with standard and non-standard markers, and a chronological analysis of marker choice among six prominent modern Chinese authors. We show that the non-standard form is more frequently used when the adverbial modifies an emotion verb. Further, we demonstrate that marker choice is correlated to text genre and register, as well as the writing style of the author.

By focusing the pre-training process on domain-specific corpora, some domain-specific pre-trained language models (PLMs) have achieved state-of-the-art results. However, it is under-investigated to design a unified paradigm to inject domain knowledge in the PLM fine-tuning stage. We propose KnowledgeDA, a unified domain language model development service to enhance the task-specific training procedure with domain knowledge graphs. Given domain-specific task texts input, KnowledgeDA can automatically generate a domain-specific language model following three steps: (i) localize domain knowledge entities in texts via an embedding-similarity approach; (ii) generate augmented samples by retrieving replaceable domain entity pairs from two views of both knowledge graph and training data; (iii) select high-quality augmented samples for fine-tuning via confidence-based assessment. We implement a prototype of KnowledgeDA to learn language models for two domains, healthcare and software development. Experiments on domain-specific text classification and QA tasks verify the effectiveness and generalizability of KnowledgeDA.

Bilingual lexicon induction induces the word translations by aligning independently trained word embeddings in two languages. Existing approaches generally focus on minimizing the distances between words in the aligned pairs, while suffering from low discriminative capability to distinguish the relative orders between positive and negative candidates. In addition, the mapping function is globally shared by all words, whose performance might be hindered by the deviations in the distributions of different languages. In this work, we propose a novel ranking-oriented induction model RAPO to learn personalized mapping function for each word. RAPO is capable of enjoying the merits from the unique characteristics of a single word and the cross-language isomorphism simultaneously. Extensive experimental results on public datasets including both rich-resource and low-resource languages demonstrate the superiority of our proposal. Our code is publicly available in https://github.com/Jlfj345wf/RAPO.

In many languages, adverbials can be derived from words of various parts-of-speech. In Chinese, the derivation may be marked either with the standard adverbial marker DI, or the non-standard marker DE. Since DE also serves double duty as the attributive marker, accurate identification of adverbials requires disambiguation of its syntactic role. As parsers are trained predominantly on texts using the standard adverbial marker DI, they often fail to recognize adverbials suffixed with the non-standard DE. This paper addresses this problem with an unsupervised, rule-based approach for adverbial identification that utilizes dependency tree patterns. Experiment results show that this approach outperforms a masked language model baseline. We apply this approach to analyze standard and non-standard adverbial marker usage in modern Chinese literature.

Recently, sponsored search has become one of the most lucrative channels for marketing. As the fundamental basis of sponsored search, relevance modeling has attracted increasing attention due to the tremendous practical value. Most existing methods solely rely on the query-keyword pairs. However, keywords are usually short texts with scarce semantic information, which may not precisely reflect the underlying advertising intents. In this paper, we investigate the novel problem of advertiser-aware relevance modeling, which leverages the advertisers’ information to bridge the gap between the search intents and advertising purposes. Our motivation lies in incorporating the unsupervised bidding behaviors as the complementary graphs to learn desirable advertiser representations. We further propose a Bidding-Graph augmented Triple-based Relevance model BGTR with three towers to deeply fuse the bidding graphs and semantic textual data. Empirically, we evaluate the BGTR model over a large industry dataset, and the experimental results consistently demonstrate its superiority.

We present Covidex, a search engine that exploits the latest neural ranking models to provide information access to the COVID-19 Open Research Dataset curated by the Allen Institute for AI. Our system has been online and serving users since late March 2020. The Covidex is the user application component of our three-pronged strategy to develop technologies for helping domain experts tackle the ongoing global pandemic. In addition, we provide robust and easy-to-use keyword search infrastructure that exploits mature fusion-based methods as well as standalone neural ranking models that can be incorporated into other applications. These techniques have been evaluated in the multi-round TREC-COVID challenge: Our infrastructure and baselines have been adopted by many participants, including some of the best systems. In round 3, we submitted the highest-scoring run that took advantage of previous training data and the second-highest fully automatic run. In rounds 4 and 5, we submitted the highest-scoring fully automatic runs.