With the advancement of technology, large language models (LLMs) have achieved remarkable performance across various natural language processing (NLP) tasks, powering LLMintegrated applications like Microsoft Copilot.However, as LLMs continue to evolve, new vulnerabilities, especially prompt injection attacks arise.These attacks trick LLMs into deviating from the original input instructions and executing the attacker's instructions injected in data content, such as retrieved results.Recent attack methods leverage LLMs' instruction-following abilities and their inabilities to distinguish instructions injected in the data content, and achieve a high attack success rate (ASR).When comparing the attack and defense methods, we interestingly find that they share similar design goals, of inducing the model to ignore unwanted instructions and instead to execute wanted instructions.Therefore, we raise an intuitive question: Could these attack techniques be utilized for defensive purposes?In this paper, we invert the intention of prompt injection methods to develop novel defense methods based on previous trainingfree attack methods, by repeating the attack process but with the original input instruction rather than the injected instruction.Our comprehensive experiments demonstrate that our defense techniques outperform existing defense approaches, achieving state-of-the-art results. 1

Previous attempts at injecting semantic frame biases into SMT training for low resource languages failed because either (a) no semantic parser is available for the low resource input language; or (b) the output English language semantic parses excise relevant parts of the alignment space too aggressively. We present the first semantic SMT model to succeed in significantly improving translation quality across many low resource input languages for which no automatic SRL is available —consistently and across all common MT metrics. The results we report are the best by far to date for this type of approach; our analyses suggest that in general, easier approaches toward including semantics in training SMT models may be more feasible than generally assumed even for low resource languages where semantic parsers remain scarce. While recent proposals to use the crosslingual evaluation metric XMEANT during inversion transduction grammar (ITG) induction are inapplicable to low resource languages that lack semantic parsers, we break the bottleneck via a vastly improved method of biasing ITG induction toward learning more semantically correct alignments using the monolingual semantic evaluation metric MEANT. Unlike XMEANT, MEANT requires only a readily-available English (output language) semantic parser. The advances we report here exploit the novel realization that MEANT represents an excellent way to semantically bias expectationmaximization induction even for low resource languages. We test our systems on challenging languages including Amharic, Uyghur, Tigrinya and Oromo. Results show that our model influences the learning towards more semantically correct alignments, leading to better translation quality than both the standard ITG or GIZA++ based SMT training models on different datasets.

We propose an approach in which we inject a crosslingual semantic frame based objective function directly into inversion transduction grammar (ITG) induction in order to semantically train spoken language translation systems. This approach represents a follow-up of our recent work on improving machine translation quality by tuning loglinear mixture weights using a semantic frame based objective function in the late, final stage of statistical machine translation training. In contrast, our new approach injects a semantic frame based objective function back into earlier stages of the training pipeline, during the actual learning of the translation model, biasing learning toward semantically more accurate alignments. Our work is motivated by the fact that ITG alignments have empirically been shown to fully cover crosslingual semantic frame alternations. We show that injecting a crosslingual semantic based objective function for driving ITG induction further sharpens the ITG constraints, leading to better performance than either the conventional ITG or the traditional GIZA++ based approaches.

We present a novel model, Freestyle, that learns to improvise rhyming and fluent responses upon being challenged with a line of hip hop lyrics, by combining both bottomup token based rule induction and top-down rule segmentation strategies to learn a stochastic transduction grammar that simultaneously learns both phrasing and rhyming associations.In this attack on the woefully under-explored natural language genre of music lyrics, we exploit a strictly unsupervised transduction grammar induction approach.Our task is particularly ambitious in that no use of any a priori linguistic or phonetic information is allowed, even though the domain of hip hop lyrics is particularly noisy and unstructured.We evaluate the performance of the learned model against a model learned only using the more conventional bottom-up token based rule induction, and demonstrate the superiority of our combined token based and rule segmentation induction method toward generating higher quality improvised responses, measured on fluency and rhyming criteria as judged by human evaluators.To highlight some of the inherent challenges in adapting other algorithms to this novel task, we also compare the quality of the responses generated by our model to those generated by an out-ofthe-box phrase based SMT system.We tackle the challenge of selecting appropriate training data for our task via a dedicated rhyme scheme detection module, which is also acquired via unsupervised learning and report improved quality of the generated responses.Finally, we report results with Maghrebi French hip hop lyrics indicating that our model performs surprisingly well with no special adaptation to other languages.

We attack an inexplicably under-explored language genre of spoken language-lyrics in music-via completely unsupervised induction of an SMT-style stochastic transduction grammar for hip hop lyrics, yielding a fully-automatically learned challenge-response system that produces rhyming lyrics given an input. Unlike previous efforts, we choose the domain of hip hop lyrics, which is particularly unstructured and noisy. A novel feature of our approach is that it is completely unsupervised and requires no a priori linguistic or phonetic knowledge. In spite of the level of difficulty of the challenge, the model nevertheless produces fluent output as judged by human evaluators, and performs significantly better than widely used phrase-based SMT models upon the same task.

We argue for an alternative paradigm in evaluating machine translation quality that is strongly empirical but more accurately reflects the utility of translations, by returning to a representational foundation based on AI oriented lexical semantics, rather than the superficial flat n-gram and string representations recently dominating the field. Driven by such metrics as BLEU and WER, current SMT frequently produces unusable translations where the semantic event structure is mistranslated: who did what to whom, when, where, why, and how? We argue that it is time for a new generation of more intelligent" automatic and semi-automatic metrics, based clearly on getting the structure right at the lexical semantics level. We show empirically that it is possible to use simple PropBank style semantic frame representations to surpass all currently widespread metrics' correlation to human adequacy judgments, including even HTER. We also show that replacing human annotators with automatic semantic role labeling still yields much of the advantage of the approach. We combine the best of both worlds: from an SMT perspective, we provide superior yet low-cost quantitative objective functions for translation quality; and yet from an AI perspective, we regain the representational transparency and clear reflection of semantic utility of structural frame-based knowledge representations.

We introduce a new type of transduction grammar that allows for learning of probabilistic phrasal bilexica, leading to a significant improvement in spoken language translation accuracy. The current state-of-the-art in statistical machine translation relies on a complicated and crude pipeline to learn probabilistic phrasal bilexica—the very core of any speech translation system. In this paper, we present a more principled approach to learning probabilistic phrasal bilexica, based on stochastic transduction grammar learning applicable to speech corpora. Index Terms: speech translation, transduction theory, lexicon extraction

In this paper, we survey some central issues in the historical, current, and future landscape of statistical machine translation (SMT) research, taking as a starting point an extended three-dimensional MT model space. We posit a socio-geographical conceptual disparity hypothesis, that aims to explain why language pairs like Chinese-English have presented MT with so much more difficulty than others. The evolution from simple token-based to segment-based to tree-based syntactic SMT is sketched. For tree-based SMT, we consider language bias rationales for selecting the degree of compositional power within the hierarchy of expressiveness for transduction grammars (or synchronous grammars). This leads us to inversion transductions and the ITG model prevalent in current state-of-the-art SMT, along with the underlying ITG hypothesis, which posits a language universal. Against this backdrop, we enumerate a set of key open questions for syntactic SMT. We then consider the more recent area of semantic SMT. We list principles for successful application of sense disambiguation models to semantic SMT, and describe early directions in the use of semantic role labeling for semantic SMT.

We present twin results on Chinese semantic parsing, with application to English-Chinese cross- lingual verb frame acquisition. First, we describe two new state-of-the-art Chinese shallow semantic parsers leading to an F-score of 82.01 on simultaneous frame and argument boundary identification and labeling. Subsequently, we propose a model that applies the separate Chinese and English semantic parsers to learn cross-lingual semantic verb frame argument mappings with 89.3% accuracy. The only training data needed by this cross-lingual learning model is a pair of non-parallel monolingual Propbanks, plus an unannotated parallel corpus. We also present the first reported controlled comparison of maximum entropy and SVM approaches to shallow semantic parsing, using the Chinese data.

We present the first known direct measurement of word alignment coverage on an Arabic-English parallel corpus using inversion transduction grammar constraints. While direct measurements have been reported for several European and Asian languages, to date no results have been available for Arabic or any Semitic language despite much recent activity on Arabic- English spoken language and text translation. Many recent syntax based statistical MT models operate within the domain of ITG expressiveness, often for efficiency reasons, so it has become important to determine the extent to which the ITG constraint assumption holds. Our results on Arabic provide further evidence that ITG expressiveness appears largely sufficient for core MT models.

We directly investigate a subject of much recent debate: do word sense disambiguation models help statistical machine translation quality? We present empirical results casting doubt on this common, but unproved, assumption. Using a state-of-the-art Chinese word sense disambiguation model to choose translation candidates for a typical IBM statistical MT system, we find that word sense disambiguation does not yield significantly better translation quality than the statistical machine translation system alone. Error analysis suggests several key factors behind this surprising finding, including inherent limitations of current statistical MT architectures.

We introduce a new method for disambiguating word senses that exploits a nonlinear Kernel Principal Component Analysis (KPCA) technique to achieve accuracy superior to the best published individual models. We present empirical results demonstrating significantly better accuracy compared to the state-of-the-art achieved by either naïve Bayes or maximum entropy models, on Senseval-2 data. We also contrast against another type of kernel method, the support vector machine (SVM) model, and show that our KPCA-based model outperforms the SVM-based model. It is hoped that these highly encouraging first results on KPCA for natural language processing tasks will inspire further development of these directions.

Portable devices such as PDA phones and smart phones are increasingly popular. Many of these devices already have voice dialing capability. The next step is to offer more powerful personal-assistant features such as speech translation. In this paper, we propose a system that can translate speech commands in Chinese into English, in real-time, on small, portable devices with limited memory and computational power. We address the various computational and platform issues of speech recognition and translation on portable devices. We propose fixed-point computation, discrete front-end speech features, bi-phone acoustic models, grammar-based speech decoding, and unambiguous inversion transduction grammars for transfer-based translation. As a result, our speech translation system requires only 500k memory and a 200MHz CPU.

The paper proposes an information-theory-based method for feature types analysis in probabilistic evaluation modelling for statistical parsing. The basic idea is that we use entropy and conditional entropy to measure whether a feature type grasps some of the information for syntactic structure prediction. Our experiment quantitatively analyzes several feature types' power for syntactic structure prediction and draws a series of interesting conclusions.

We investigate the utility of right-context #lookahead information# in incremental left-to-right language models with word sense disambiguation, and discover somewhat unexpectedly that using right-context in addition to left-context #history# may actually reduce accuracy.

We depict the interpolation strategies for the concatenation of inventory demisyllables in our hybrid concatenative/LPC speech synthesizer. Inventory elements for vowels and nasals are cut in the steady state of the phoneme. Concatenating elements in the synthesis stage requires smoothing of spectral content and energy to avoid annoying discontinuities in these parameters, which is of vital importance for the quality of synthesized speech. The hybrid synthesizer concept allows the application of smoothing algorithms uncommon to pure time domain synthesis. Using the smoothing algorithm also the occurrence of musical tones due to the LPC filter resonances at the begin of pauses can be suppressed.

In this paper, a distance-related unit association maximum entropy (DUAME) language modeling is proposed. This approach can model an event (unit subsequence) using the co-occurrence of full distance unit association (UA) features so that it is able to pursue a functional approximation to higher order N-gram with significantly less memory requirement. A smoothing strategy related to this modeling will also be discussed. Preliminary experimental results have shown that DUAME modeling is comparable to conventional N-gram modeling in perplexity with significantly small number of parameters.

We introduce a polynomial-time algorithm for statistical machine translation. This algorithm can be used in place of the expensive, slow best-first search strategies in current statistical translation architectures. The approach employs the stochastic bracketing transduction grammar (SBTG) model we recently introduced to replace earlier word alignment channel models, while retaining a bigram language model. The new algorithm in our experience yields major speed improvement with no significant loss of accuracy.

We describe a grammarless method for simultaneously bracketing both halves of a parallel text and giving word alignments, assuming only a translation lexicon for the language pair. We introduce inversion-invariant transduction grammars which serve as generative models for parallel bilingual sentences with weak order constraints. Focusing on transduction grammars for bracketing, we formulate a normal form, and a stochastic version amenable to a maximum-likelihood bracketing algorithm. Several extensions and experiments are discussed.

We describe our experience with automatic alignment of sentences in parallel English-Chinese texts. Our report concerns three related topics: (1) progress on the HKUST English-Chinese Parallel Bilingual Corpus; (2) experiments addressing the applicability of Gale & Church's (1991) length-based statistical method to the task of alignment involving a non-Indo-European language; and (3) an improved statistical method that also incorporates domain-specific lexical cues.