Alex Waibel

Ngoc-Quan Pham, Alex Waibel, Jan Niehues

Multilingual speech recognition with supervised learning has achieved great results as reflected in recent research. With the development of pretraining methods on audio and text data, it is imperative to transfer the knowledge from unsupervised multilingual models to facilitate recognition, especially in many languages with limited data. Our work investigated the effectiveness of using two pretrained models for two modalities: wav2vec 2.0 for audio and MBART50 for text, together with the adaptive weight techniques to massively improve the recognition quality on the public datasets containing CommonVoice and Europarl. Overall, we noticed an 44% improvement over purely supervised learning, and more importantly, each technique provides a different reinforcement in different languages. We also explore other possibilities to potentially obtain the best model by slightly adding either depth or relative attention to the architecture.

Leonard Bärmann, Rainer Kartmann, Fabian Peller-Konrad et al.

Natural-language dialog is key for intuitive human-robot interaction. It can be used not only to express humans' intents, but also to communicate instructions for improvement if a robot does not understand a command correctly. Of great importance is to endow robots with the ability to learn from such interaction experience in an incremental way to allow them to improve their behaviors or avoid mistakes in the future. In this paper, we propose a system to achieve incremental learning of complex behavior from natural interaction, and demonstrate its implementation on a humanoid robot. Building on recent advances, we present a system that deploys Large Language Models (LLMs) for high-level orchestration of the robot's behavior, based on the idea of enabling the LLM to generate Python statements in an interactive console to invoke both robot perception and action. The interaction loop is closed by feeding back human instructions, environment observations, and execution results to the LLM, thus informing the generation of the next statement. Specifically, we introduce incremental prompt learning, which enables the system to interactively learn from its mistakes. For that purpose, the LLM can call another LLM responsible for code-level improvements of the current interaction based on human feedback. The improved interaction is then saved in the robot's memory, and thus retrieved on similar requests. We integrate the system in the robot cognitive architecture of the humanoid robot ARMAR-6 and evaluate our methods both quantitatively (in simulation) and qualitatively (in simulation and real-world) by demonstrating generalized incrementally-learned knowledge.

Peter Polák, Brian Yan, Shinji Watanabe et al.

Blockwise self-attentional encoder models have recently emerged as one promising end-to-end approach to simultaneous speech translation. These models employ a blockwise beam search with hypothesis reliability scoring to determine when to wait for more input speech before translating further. However, this method maintains multiple hypotheses until the entire speech input is consumed -- this scheme cannot directly show a single \textit{incremental} translation to users. Further, this method lacks mechanisms for \textit{controlling} the quality vs. latency tradeoff. We propose a modified incremental blockwise beam search incorporating local agreement or hold-$n$ policies for quality-latency control. We apply our framework to models trained for online or offline translation and demonstrate that both types can be effectively used in online mode. Experimental results on MuST-C show 0.6-3.6 BLEU improvement without changing latency or 0.8-1.4 s latency improvement without changing quality.

Leonard Bärmann, Chad DeChant, Joana Plewnia et al.

Verbalization of robot experience, i.e., summarization of and question answering about a robot's past, is a crucial ability for improving human-robot interaction. Previous works applied rule-based systems or fine-tuned deep models to verbalize short (several-minute-long) streams of episodic data, limiting generalization and transferability. In our work, we apply large pretrained models to tackle this task with zero or few examples, and specifically focus on verbalizing life-long experiences. For this, we derive a tree-like data structure from episodic memory (EM), with lower levels representing raw perception and proprioception data, and higher levels abstracting events to natural language concepts. Given such a hierarchical representation built from the experience stream, we apply a large language model as an agent to interactively search the EM given a user's query, dynamically expanding (initially collapsed) tree nodes to find the relevant information. The approach keeps computational costs low even when scaling to months of robot experience data. We evaluate our method on simulated household robot data, human egocentric videos, and real-world robot recordings, demonstrating its flexibility and scalability.

Leonard Bärmann, Joana Plewnia, Alex Waibel et al.

Robots must verbalize their past experiences when users ask "Where did you put my keys?" or "Why did the task fail?" Yet maintaining life-long episodic memory (EM) from continuous multimodal perception quickly exceeds storage limits and makes real-time query impractical, calling for selective forgetting that adapts to users' notions of relevance. We present H$^2$-EMV, a framework enabling humanoids to learn what to remember through user interaction. Our approach incrementally constructs hierarchical EM, selectively forgets using language-model-based relevance estimation conditioned on learned natural-language rules, and updates these rules given user feedback about forgotten details. Evaluations on simulated household tasks and 20.5-hour-long real-world recordings from ARMAR-7 demonstrate that H$^2$-EMV maintains question-answering accuracy while reducing memory size by 45% and query-time compute by 35%. Critically, performance improves over time - accuracy increases 70% in second-round queries by adapting to user-specific priorities - demonstrating that learned forgetting enables scalable, personalized EM for long-term human-robot collaboration.

Ibrahim Said Ahmad, Antonios Anastasopoulos, Ondřej Bojar et al.

This paper reports on the shared tasks organized by the 21st IWSLT Conference. The shared tasks address 7 scientific challenges in spoken language translation: simultaneous and offline translation, automatic subtitling and dubbing, speech-to-speech translation, dialect and low-resource speech translation, and Indic languages. The shared tasks attracted 18 teams whose submissions are documented in 26 system papers. The growing interest towards spoken language translation is also witnessed by the constantly increasing number of shared task organizers and contributors to the overview paper, almost evenly distributed across industry and academia.

Enes Yavuz Ugan, Ngoc-Quan Pham, Leonard Bärmann et al.

Code-switching, the alternation of languages within a single discourse, presents a significant challenge for Automatic Speech Recognition. Despite the unique nature of the task, performance is commonly measured with established metrics such as Word-Error-Rate (WER). However, in this paper, we question whether these general metrics accurately assess performance on code-switching. Specifically, using both Connectionist-Temporal-Classification and Encoder-Decoder models, we show fine-tuning on non-code-switched data from both matrix and embedded language improves classical metrics on code-switching test sets, although actual code-switched words worsen (as expected). Therefore, we propose Point-of-Interest Error Rate (PIER), a variant of WER that focuses only on specific words of interest. We instantiate PIER on code-switched utterances and show that this more accurately describes the code-switching performance, showing huge room for improvement in future work. This focused evaluation allows for a more precise assessment of model performance, particularly in challenging aspects such as inter-word and intra-word code-switching.

Felix Schneider, Marco Turchi, Alex Waibel

With more and more meetings moving to a digital domain, meeting summarization has recently gained interest in both academic and commercial research. However, prior academic research focuses on meeting summarization as an offline task, performed after the meeting concludes. In this paper, we perform the first systematic study of online meeting summarization. For this purpose, we propose several policies for conducting online summarization. We discuss the unique challenges of this task compared to the offline setting and define novel metrics to evaluate latency and partial summary quality. The experiments on the AutoMin dataset show that 1) online models can produce strong summaries, 2) our metrics allow a detailed analysis of different systems' quality-latency trade-off, also taking into account intermediate outputs and 3) adaptive policies perform better than fixed scheduled ones. These findings provide a starting point for the wider research community to explore this important task.

Zhong Zhou, Jan Niehues, Alex Waibel

In many humanitarian scenarios, translation into severely low resource languages often does not require a universal translation engine, but a dedicated text-specific translation engine. For example, healthcare records, hygienic procedures, government communication, emergency procedures and religious texts are all limited texts. While generic translation engines for all languages do not exist, translation of multilingually known limited texts into new, endangered languages may be possible and reduce human translation effort. We attempt to leverage translation resources from many rich resource languages to efficiently produce best possible translation quality for a well known text, which is available in multiple languages, in a new, severely low resource language. We examine two approaches: 1. best selection of seed sentences to jump start translations in a new language in view of best generalization to the remainder of a larger targeted text(s), and 2. we adapt large general multilingual translation engines from many other languages to focus on a specific text in a new, unknown language. We find that adapting large pretrained multilingual models to the domain/text first and then to the severely low resource language works best. If we also select a best set of seed sentences, we can improve average chrF performance on new test languages from a baseline of 21.9 to 50.7, while reducing the number of seed sentences to only around 1,000 in the new, unknown language.

Zhong Zhou, Alex Waibel

We translate a closed text that is known in advance and available in many languages into a new and severely low resource language. Most human translation efforts adopt a portion-based approach to translate consecutive pages/chapters in order, which may not suit machine translation. We compare the portion-based approach that optimizes coherence of the text locally with the random sampling approach that increases coverage of the text globally. Our results show that the random sampling approach performs better. When training on a seed corpus of ~1,000 lines from the Bible and testing on the rest of the Bible (~30,000 lines), random sampling gives a performance gain of +11.0 BLEU using English as a simulated low resource language, and +4.9 BLEU using Eastern Pokomchi, a Mayan language. Furthermore, we compare three ways of updating machine translation models with increasing amount of human post-edited data through iterations. We find that adding newly post-edited data to training after vocabulary update without self-supervision performs the best. We propose an algorithm for human and machine to work together seamlessly to translate a closed text into a severely low resource language.

Zhong Zhou, Alex Waibel

We translate a closed text that is known in advance into a severely low resource language by leveraging massive source parallelism. In other words, given a text in 124 source languages, we translate it into a severely low resource language using only ~1,000 lines of low resource data without any external help. Firstly, we propose a systematic method to rank and choose source languages that are close to the low resource language. We call the linguistic definition of language family Family of Origin (FAMO), and we call the empirical definition of higher-ranked languages using our metrics Family of Choice (FAMC). Secondly, we build an Iteratively Pretrained Multilingual Order-preserving Lexiconized Transformer (IPML) to train on ~1,000 lines (~3.5%) of low resource data. To translate named entities correctly, we build a massive lexicon table for 2,939 Bible named entities in 124 source languages, and include many that occur once and covers more than 66 severely low resource languages. Moreover, we also build a novel method of combining translations from different source languages into one. Using English as a hypothetical low resource language, we get a +23.9 BLEU increase over a multilingual baseline, and a +10.3 BLEU increase over our asymmetric baseline in the Bible dataset. We get a 42.8 BLEU score for Portuguese-English translation on the medical EMEA dataset. We also have good results for a real severely low resource Mayan language, Eastern Pokomchi.

Thai-Son Nguyen, Sebastian Stueker, Alex Waibel

Achieving super-human performance in recognizing human speech has been a goal for several decades, as researchers have worked on increasingly challenging tasks. In the 1990's it was discovered, that conversational speech between two humans turns out to be considerably more difficult than read speech as hesitations, disfluencies, false starts and sloppy articulation complicate acoustic processing and require robust handling of acoustic, lexical and language context, jointly. Early attempts with statistical models could only reach error rates over 50% and far from human performance (WER of around 5.5%). Neural hybrid models and recent attention-based encoder-decoder models have considerably improved performance as such contexts can now be learned in an integral fashion. However, processing such contexts requires an entire utterance presentation and thus introduces unwanted delays before a recognition result can be output. In this paper, we address performance as well as latency. We present results for a system that can achieve super-human performance (at a WER of 5.0%, over the Switchboard conversational benchmark) at a word based latency of only 1 second behind a speaker's speech. The system uses multiple attention-based encoder-decoder networks integrated within a novel low latency incremental inference approach.

Stefan Constantin, Alex Waibel

We propose a dialog system utility component that gets the last two utterances of a user and can detect whether the last utterance is an error correction of the second last utterance. If yes, it corrects the second last utterance according to the error correction in the last utterance and outputs the extracted pairs of reparandum and repair entity. This component offers two advantages, learning the concept of corrections to avoid collecting corrections for every new domain and extracting reparandum and repair pairs, which offers the possibility to learn out of it. For the error correction one sequence labeling and two sequence to sequence approaches are presented. For the error correction detection these three error correction approaches can also be used and in addition, we present a sequence classification approach. One error correction detection and one error correction approach can be combined to a pipeline or the error correction approaches can be trained and used end-to-end to avoid two components. We modified the EPIC-KITCHENS-100 dataset to evaluate the approaches for correcting entity phrases in request dialogs. For error correction detection and correction, we got an accuracy of 96.40 % on synthetic validation data and an accuracy of 77.81 % on human-created real-world test data.

Thai-Son Nguyen, Ngoc-Quan Pham, Sebastian Stueker et al.

Recently sequence-to-sequence models have started to achieve state-of-the-art performance on standard speech recognition tasks when processing audio data in batch mode, i.e., the complete audio data is available when starting processing. However, when it comes to performing run-on recognition on an input stream of audio data while producing recognition results in real-time and with low word-based latency, these models face several challenges. For many techniques, the whole audio sequence to be decoded needs to be available at the start of the processing, e.g., for the attention mechanism or the bidirectional LSTM (BLSTM). In this paper, we propose several techniques to mitigate these problems. We introduce an additional loss function controlling the uncertainty of the attention mechanism, a modified beam search identifying partial, stable hypotheses, ways of working with BLSTM in the encoder, and the use of chunked BLSTM. Our experiments show that with the right combination of these techniques, it is possible to perform run-on speech recognition with low word-based latency without sacrificing in word error rate performance.

Thai Son Nguyen, Jan Niehues, Eunah Cho et al.

User studies have shown that reducing the latency of our simultaneous lecture translation system should be the most important goal. We therefore have worked on several techniques for reducing the latency for both components, the automatic speech recognition and the speech translation module. Since the commonly used commitment latency is not appropriate in our case of continuous stream decoding, we focused on word latency. We used it to analyze the performance of our current system and to identify opportunities for improvements. In order to minimize the latency we combined run-on decoding with a technique for identifying stable partial hypotheses when stream decoding and a protocol for dynamic output update that allows to revise the most recent parts of the transcription. This combination reduces the latency at word level, where the words are final and will never be updated again in the future, from 18.1s to 1.1s without sacrificing performance in terms of word error rate.

Thai-Son Nguyen, Sebastian Stüker, Alex Waibel

In the area of multi-domain speech recognition, research in the past focused on hybrid acoustic models to build cross-domain and domain-invariant speech recognition systems. In this paper, we empirically examine the difference in behavior between hybrid acoustic models and neural end-to-end systems when mixing acoustic training data from several domains. For these experiments we composed a multi-domain dataset from public sources, with the different domains in the corpus covering a wide variety of topics and acoustic conditions such as telephone conversations, lectures, read speech and broadcast news. We show that for the hybrid models, supplying additional training data from other domains with mismatched acoustic conditions does not increase the performance on specific domains. However, our end-to-end models optimized with sequence-based criterion generalize better than the hybrid models on diverse domains. In term of word-error-rate performance, our experimental acoustic-to-word and attention-based models trained on multi-domain dataset reach the performance of domain-specific long short-term memory (LSTM) hybrid models, thus resulting in multi-domain speech recognition systems that do not suffer in performance over domain specific ones. Moreover, the use of neural end-to-end models eliminates the need of domain-adapted language models during recognition, which is a great advantage when the input domain is unknown.

Tino Fuhrman, David Schneider, Felix Altenberg et al.

With the rapid advance of sophisticated control algorithms, the capabilities of drones to stabilise, fly and manoeuvre autonomously have dramatically improved, enabling us to pay greater attention to entire missions and the interaction of a drone with humans and with its environment during the course of such a mission. In this paper, we present an indoor office drone assistant that is tasked to run errands and carry out simple tasks at our laboratory, while given instructions from and interacting with humans in the space. To accomplish its mission, the system has to be able to understand verbal instructions from humans, and perform subject to constraints from control and hardware limitations, uncertain localisation information, unpredictable and uncertain obstacles and environmental factors. We combine and evaluate the dialogue, navigation, flight control, depth perception and collision avoidance components. We discuss performance and limitations of our assistant at the component as well as the mission level. A 78% mission success rate was obtained over the course of 27 missions.

Verena Heusser, Niklas Freymuth, Stefan Constantin et al.

Speech emotion recognition is a challenging task and an important step towards more natural human-machine interaction. We show that pre-trained language models can be fine-tuned for text emotion recognition, achieving an accuracy of 69.5% on Task 4A of SemEval 2017, improving upon the previous state of the art by over 3% absolute. We combine these language models with speech emotion recognition, achieving results of 73.5% accuracy when using provided transcriptions and speech data on a subset of four classes of the IEMOCAP dataset. The use of noise-induced transcriptions and speech data results in an accuracy of 71.4%. For our experiments, we created IEmoNet, a modular and adaptable bimodal framework for speech emotion recognition based on pre-trained language models. Lastly, we discuss the idea of using an emotional classifier as a reward for reinforcement learning as a step towards more successful and convenient human-machine interaction.

Zhong Zhou, Lori Levin, David R. Mortensen et al.

We demonstrate a new approach to Neural Machine Translation (NMT) for low-resource languages using a ubiquitous linguistic resource, Interlinear Glossed Text (IGT). IGT represents a non-English sentence as a sequence of English lemmas and morpheme labels. As such, it can serve as a pivot or interlingua for NMT. Our contribution is four-fold. Firstly, we pool IGT for 1,497 languages in ODIN (54,545 glosses) and 70,918 glosses in Arapaho and train a gloss-to-target NMT system from IGT to English, with a BLEU score of 25.94. We introduce a multilingual NMT model that tags all glossed text with gloss-source language tags and train a universal system with shared attention across 1,497 languages. Secondly, we use the IGT gloss-to-target translation as a key step in an English-Turkish MT system trained on only 865 lines from ODIN. Thirdly, we we present five metrics for evaluating extremely low-resource translation when BLEU is no longer sufficient and evaluate the Turkish low-resource system using BLEU and also using accuracy of matching nouns, verbs, agreement, tense, and spurious repetition, showing large improvements.

Thai-Son Nguyen, Sebastian Stueker, Jan Niehues et al.

Sequence-to-Sequence (S2S) models recently started to show state-of-the-art performance for automatic speech recognition (ASR). With these large and deep models overfitting remains the largest problem, outweighing performance improvements that can be obtained from better architectures. One solution to the overfitting problem is increasing the amount of available training data and the variety exhibited by the training data with the help of data augmentation. In this paper we examine the influence of three data augmentation methods on the performance of two S2S model architectures. One of the data augmentation method comes from literature, while two other methods are our own development - a time perturbation in the frequency domain and sub-sequence sampling. Our experiments on Switchboard and Fisher data show state-of-the-art performance for S2S models that are trained solely on the speech training data and do not use additional text data.

Stefan Constantin, Jan Niehues, Alex Waibel

The state-of-the-art neural network architectures make it possible to create spoken language understanding systems with high quality and fast processing time. One major challenge for real-world applications is the high latency of these systems caused by triggered actions with high executions times. If an action can be separated into subactions, the reaction time of the systems can be improved through incremental processing of the user utterance and starting subactions while the utterance is still being uttered. In this work, we present a model-agnostic method to achieve high quality in processing incrementally produced partial utterances. Based on clean and noisy versions of the ATIS dataset, we show how to create datasets with our method to create low-latency natural language understanding components. We get improvements of up to 47.91 absolute percentage points in the metric F1-score.

Ngoc-Quan Pham, Jan Niehues, Thanh-Le Ha et al.

An important concern in training multilingual neural machine translation (NMT) is to translate between language pairs unseen during training, i.e zero-shot translation. Improving this ability kills two birds with one stone by providing an alternative to pivot translation which also allows us to better understand how the model captures information between languages. In this work, we carried out an investigation on this capability of the multilingual NMT models. First, we intentionally create an encoder architecture which is independent with respect to the source language. Such experiments shed light on the ability of NMT encoders to learn multilingual representations, in general. Based on such proof of concept, we were able to design regularization methods into the standard Transformer model, so that the whole architecture becomes more robust in zero-shot conditions. We investigated the behaviour of such models on the standard IWSLT 2017 multilingual dataset. We achieved an average improvement of 2.23 BLEU points across 12 language pairs compared to the zero-shot performance of a state-of-the-art multilingual system. Additionally, we carry out further experiments in which the effect is confirmed even for language pairs with multiple intermediate pivots.

Matthias Sperber, Graham Neubig, Ngoc-Quan Pham et al.

Lattices are an efficient and effective method to encode ambiguity of upstream systems in natural language processing tasks, for example to compactly capture multiple speech recognition hypotheses, or to represent multiple linguistic analyses. Previous work has extended recurrent neural networks to model lattice inputs and achieved improvements in various tasks, but these models suffer from very slow computation speeds. This paper extends the recently proposed paradigm of self-attention to handle lattice inputs. Self-attention is a sequence modeling technique that relates inputs to one another by computing pairwise similarities and has gained popularity for both its strong results and its computational efficiency. To extend such models to handle lattices, we introduce probabilistic reachability masks that incorporate lattice structure into the model and support lattice scores if available. We also propose a method for adapting positional embeddings to lattice structures. We apply the proposed model to a speech translation task and find that it outperforms all examined baselines while being much faster to compute than previous neural lattice models during both training and inference.

Elizabeth Salesky, Matthias Sperber, Alex Waibel

Spoken language translation applications for speech suffer due to conversational speech phenomena, particularly the presence of disfluencies. With the rise of end-to-end speech translation models, processing steps such as disfluency removal that were previously an intermediate step between speech recognition and machine translation need to be incorporated into model architectures. We use a sequence-to-sequence model to translate from noisy, disfluent speech to fluent text with disfluencies removed using the recently collected `copy-edited' references for the Fisher Spanish-English dataset. We are able to directly generate fluent translations and introduce considerations about how to evaluate success on this task. This work provides a baseline for a new task, the translation of conversational speech with joint removal of disfluencies.

Matthias Sperber, Graham Neubig, Jan Niehues et al.

Speech translation has traditionally been approached through cascaded models consisting of a speech recognizer trained on a corpus of transcribed speech, and a machine translation system trained on parallel texts. Several recent works have shown the feasibility of collapsing the cascade into a single, direct model that can be trained in an end-to-end fashion on a corpus of translated speech. However, experiments are inconclusive on whether the cascade or the direct model is stronger, and have only been conducted under the unrealistic assumption that both are trained on equal amounts of data, ignoring other available speech recognition and machine translation corpora. In this paper, we demonstrate that direct speech translation models require more data to perform well than cascaded models, and while they allow including auxiliary data through multi-task training, they are poor at exploiting such data, putting them at a severe disadvantage. As a remedy, we propose the use of end-to-end trainable models with two attention mechanisms, the first establishing source speech to source text alignments, the second modeling source to target text alignment. We show that such models naturally decompose into multi-task-trainable recognition and translation tasks and propose an attention-passing technique that alleviates error propagation issues in a previous formulation of a model with two attention stages. Our proposed model outperforms all examined baselines and is able to exploit auxiliary training data much more effectively than direct attentional models.

Thai-Son Nguyen, Sebastian Stueker, Alex Waibel

In this work, we learn a shared encoding representation for a multi-task neural network model optimized with connectionist temporal classification (CTC) and conventional framewise cross-entropy training criteria. Our experiments show that the multi-task training not only tackles the complexity of optimizing CTC models such as acoustic-to-word but also results in significant improvement compared to the plain-task training with an optimal setup. Furthermore, we propose to use the encoding representation learned by the multi-task network to initialize the encoder of attention-based models. Thereby, we train a deep attention-based end-to-end model with 10 long short-term memory (LSTM) layers of encoder which produces 12.2\% and 22.6\% word-error-rate on Switchboard and CallHome subsets of the Hub5 2000 evaluation.

Thai-Son Nguyen, Sebastian Stueker, Alex Waibel

Acoustic-to-word (A2W) models that allow direct mapping from acoustic signals to word sequences are an appealing approach to end-to-end automatic speech recognition due to their simplicity. However, prior works have shown that modelling A2W typically encounters issues of data sparsity that prevent training such a model directly. So far, pre-training initialization is the only approach proposed to deal with this issue. In this work, we propose to build a shared neural network and optimize A2W and conventional hybrid models in a multi-task manner. Our results show that training an A2W model is much more stable with our multi-task model without pre-training initialization, and results in a significant improvement compared to a baseline model. Experiments also reveal that the performance of a hybrid acoustic model can be further improved when jointly training with a sequence-level optimization criterion such as acoustic-to-word.

Stefan Constantin, Jan Niehues, Alex Waibel

Recently advancements in sequence-to-sequence neural network architectures have led to an improved natural language understanding. When building a neural network-based Natural Language Understanding component, one main challenge is to collect enough training data. The generation of a synthetic dataset is an inexpensive and quick way to collect data. Since this data often has less variety than real natural language, neural networks often have problems to generalize to unseen utterances during testing. In this work, we address this challenge by using multi-task learning. We train out-of-domain real data alongside in-domain synthetic data to improve natural language understanding. We evaluate this approach in the domain of airline travel information with two synthetic datasets. As out-of-domain real data, we test two datasets based on the subtitles of movies and series. By using an attention-based encoder-decoder model, we were able to improve the F1-score over strong baselines from 80.76 % to 84.98 % in the smaller synthetic dataset.

Elizabeth Salesky, Susanne Burger, Jan Niehues et al.

When translating from speech, special consideration for conversational speech phenomena such as disfluencies is necessary. Most machine translation training data consists of well-formed written texts, causing issues when translating spontaneous speech. Previous work has introduced an intermediate step between speech recognition (ASR) and machine translation (MT) to remove disfluencies, making the data better-matched to typical translation text and significantly improving performance. However, with the rise of end-to-end speech translation systems, this intermediate step must be incorporated into the sequence-to-sequence architecture. Further, though translated speech datasets exist, they are typically news or rehearsed speech without many disfluencies (e.g. TED), or the disfluencies are translated into the references (e.g. Fisher). To generate clean translations from disfluent speech, cleaned references are necessary for evaluation. We introduce a corpus of cleaned target data for the Fisher Spanish-English dataset for this task. We compare how different architectures handle disfluencies and provide a baseline for removing disfluencies in end-to-end translation.

Ngoc-Quan Pham, Jan Niehues, Alex Waibel

Neural machine translation (NMT) has significantly improved the quality of automatic translation models. One of the main challenges in current systems is the translation of rare words. We present a generic approach to address this weakness by having external models annotate the training data as Experts, and control the model-expert interaction with a pointer network and reinforcement learning. Our experiments using phrase-based models to simulate Experts to complement neural machine translation models show that the model can be trained to copy the annotations into the output consistently. We demonstrate the benefit of our proposed framework in outof-domain translation scenarios with only lexical resources, improving more than 1.0 BLEU point in both translation directions English to Spanish and German to English

Zhong Zhou, Matthias Sperber, Alex Waibel

Paraphrases, the rewordings of the same semantic meaning, are useful for improving generalization and translation. However, prior works only explore paraphrases at the word or phrase level, not at the sentence or corpus level. Unlike previous works that only explore paraphrases at the word or phrase level, we use different translations of the whole training data that are consistent in structure as paraphrases at the corpus level. We train on parallel paraphrases in multiple languages from various sources. We treat paraphrases as foreign languages, tag source sentences with paraphrase labels, and train on parallel paraphrases in the style of multilingual Neural Machine Translation (NMT). Our multi-paraphrase NMT that trains only on two languages outperforms the multilingual baselines. Adding paraphrases improves the rare word translation and increases entropy and diversity in lexical choice. Adding the source paraphrases boosts performance better than adding the target ones. Combining both the source and the target paraphrases lifts performance further; combining paraphrases with multilingual data helps but has mixed performance. We achieve a BLEU score of 57.2 for French-to-English translation using 24 corpus-level paraphrases of the Bible, which outperforms the multilingual baselines and is +34.7 above the single-source single-target NMT baseline.

Jan Niehues, Ngoc-Quan Pham, Thanh-Le Ha et al.

Through the development of neural machine translation, the quality of machine translation systems has been improved significantly. By exploiting advancements in deep learning, systems are now able to better approximate the complex mapping from source sentences to target sentences. But with this ability, new challenges also arise. An example is the translation of partial sentences in low-latency speech translation. Since the model has only seen complete sentences in training, it will always try to generate a complete sentence, though the input may only be a partial sentence. We show that NMT systems can be adapted to scenarios where no task-specific training data is available. Furthermore, this is possible without losing performance on the original training data. We achieve this by creating artificial data and by using multi-task learning. After adaptation, we are able to reduce the number of corrections displayed during incremental output construction by 45%, without a decrease in translation quality.

Patrick Huber, Jan Niehues, Alex Waibel

We present a new recurrent neural network topology to enhance state-of-the-art machine learning systems by incorporating a broader context. Our approach overcomes recent limitations with extended narratives through a multi-layered computational approach to generate an abstract context representation. Therefore, the developed system captures the narrative on word-level, sentence-level, and context-level. Through the hierarchical set-up, our proposed model summarizes the most salient information on each level and creates an abstract representation of the extended context. We subsequently use this representation to enhance neural language processing systems on the task of semantic error detection. To show the potential of the newly introduced topology, we compare the approach against a context-agnostic set-up including a standard neural language model and a supervised binary classification network. The performance measures on the error detection task show the advantage of the hierarchical context-aware topologies, improving the baseline by 12.75% relative for unsupervised models and 20.37% relative for supervised models.

Jörg Franke, Jan Niehues, Alex Waibel

Deep learning models are often not easily adaptable to new tasks and require task-specific adjustments. The differentiable neural computer (DNC), a memory-augmented neural network, is designed as a general problem solver which can be used in a wide range of tasks. But in reality, it is hard to apply this model to new tasks. We analyze the DNC and identify possible improvements within the application of question answering. This motivates a more robust and scalable DNC (rsDNC). The objective precondition is to keep the general character of this model intact while making its application more reliable and speeding up its required training time. The rsDNC is distinguished by a more robust training, a slim memory unit and a bidirectional architecture. We not only achieve new state-of-the-art performance on the bAbI task, but also minimize the performance variance between different initializations. Furthermore, we demonstrate the simplified applicability of the rsDNC to new tasks with passable results on the CNN RC task without adaptions.

Markus Müller, Sebastian Stüker, Alex Waibel

Multilingual Speech Recognition is one of the most costly AI problems, because each language (7,000+) and even different accents require their own acoustic models to obtain best recognition performance. Even though they all use the same phoneme symbols, each language and accent imposes its own coloring or "twang". Many adaptive approaches have been proposed, but they require further training, additional data and generally are inferior to monolingually trained models. In this paper, we propose a different approach that uses a large multilingual model that is \emph{modulated} by the codes generated by an ancillary network that learns to code useful differences between the "twangs" or human language. We use Meta-Pi networks to have one network (the language code net) gate the activity of neurons in another (the acoustic model nets). Our results show that during recognition multilingual Meta-Pi networks quickly adapt to the proper language coloring without retraining or new data, and perform better than monolingually trained networks. The model was evaluated by training acoustic modeling nets and modulating language code nets jointly and optimize them for best recognition performance.

Zhong Zhou, Matthias Sperber, Alex Waibel

We work on translation from rich-resource languages to low-resource languages. The main challenges we identify are the lack of low-resource language data, effective methods for cross-lingual transfer, and the variable-binding problem that is common in neural systems. We build a translation system that addresses these challenges using eight European language families as our test ground. Firstly, we add the source and the target family labels and study intra-family and inter-family influences for effective cross-lingual transfer. We achieve an improvement of +9.9 in BLEU score for English-Swedish translation using eight families compared to the single-family multi-source multi-target baseline. Moreover, we find that training on two neighboring families closest to the low-resource language is often enough. Secondly, we construct an ablation study and find that reasonably good results can be achieved even with considerably less target data. Thirdly, we address the variable-binding problem by building an order-preserving named entity translation model. We obtain 60.6% accuracy in qualitative evaluation where our translations are akin to human translations in a preliminary study.

Matthias Sperber, Jan Niehues, Graham Neubig et al.

Self-attention is a method of encoding sequences of vectors by relating these vectors to each-other based on pairwise similarities. These models have recently shown promising results for modeling discrete sequences, but they are non-trivial to apply to acoustic modeling due to computational and modeling issues. In this paper, we apply self-attention to acoustic modeling, proposing several improvements to mitigate these issues: First, self-attention memory grows quadratically in the sequence length, which we address through a downsampling technique. Second, we find that previous approaches to incorporate position information into the model are unsuitable and explore other representations and hybrid models to this end. Third, to stress the importance of local context in the acoustic signal, we propose a Gaussian biasing approach that allows explicit control over the context range. Experiments find that our model approaches a strong baseline based on LSTMs with network-in-network connections while being much faster to compute. Besides speed, we find that interpretability is a strength of self-attentional acoustic models, and demonstrate that self-attention heads learn a linguistically plausible division of labor.

Patrick Huber, Jan Niehues, Alex Waibel

We present a new approach to evaluate computational models for the task of text understanding by the means of out-of-context error detection. Through the novel design of our automated modification process, existing large-scale data sources can be adopted for a vast number of text understanding tasks. The data is thereby altered on a semantic level, allowing models to be tested against a challenging set of modified text passages that require to comprise a broader narrative discourse. Our newly introduced task targets actual real-world problems of transcription and translation systems by inserting authentic out-of-context errors. The automated modification process is applied to the 2016 TEDTalk corpus. Entirely automating the process allows the adoption of complete datasets at low cost, facilitating supervised learning procedures and deeper networks to be trained and tested. To evaluate the quality of the modification algorithm a language model and a supervised binary classification model are trained and tested on the altered dataset. A human baseline evaluation is examined to compare the results with human performance. The outcome of the evaluation task indicates the difficulty to detect semantic errors for machine-learning algorithms and humans, showing that the errors cannot be identified when limited to a single sentence.

Stefan Constantin, Jan Niehues, Alex Waibel

Recently advancements in deep learning allowed the development of end-to-end trained goal-oriented dialog systems. Although these systems already achieve good performance, some simplifications limit their usage in real-life scenarios. In this work, we address two of these limitations: ignoring positional information and a fixed number of possible response candidates. We propose to use positional encodings in the input to model the word order of the user utterances. Furthermore, by using a feedforward neural network, we are able to generate the output word by word and are no longer restricted to a fixed number of possible response candidates. Using the positional encoding, we were able to achieve better accuracies in the Dialog bAbI Tasks and using the feedforward neural network for generating the response, we were able to save computation time and space consumption.

Thomas Zenkel, Ramon Sanabria, Florian Metze et al.

This paper proposes a novel approach to create an unit set for CTC based speech recognition systems. By using Byte Pair Encoding we learn an unit set of an arbitrary size on a given training text. In contrast to using characters or words as units this allows us to find a good trade-off between the size of our unit set and the available training data. We evaluate both Crossword units, that may span multiple word, and Subword units. By combining this approach with decoding methods using a separate language model we are able to achieve state of the art results for grapheme based CTC systems.

Markus Müller, Sebastian Stüker, Alex Waibel

In this work, we focus on multilingual systems based on recurrent neural networks (RNNs), trained using the Connectionist Temporal Classification (CTC) loss function. Using a multilingual set of acoustic units poses difficulties. To address this issue, we proposed Language Feature Vectors (LFVs) to train language adaptive multilingual systems. Language adaptation, in contrast to speaker adaptation, needs to be applied not only on the feature level, but also to deeper layers of the network. In this work, we therefore extended our previous approach by introducing a novel technique which we call "modulation". Based on this method, we modulated the hidden layers of RNNs using LFVs. We evaluated this approach in both full and low resource conditions, as well as for grapheme and phone based systems. Lower error rates throughout the different conditions could be achieved by the use of the modulation.

Markus Müller, Sebastian Stüker, Alex Waibel

Training automatic speech recognition (ASR) systems requires large amounts of data in the target language in order to achieve good performance. Whereas large training corpora are readily available for languages like English, there exists a long tail of languages which do suffer from a lack of resources. One method to handle data sparsity is to use data from additional source languages and build a multilingual system. Recently, ASR systems based on recurrent neural networks (RNNs) trained with connectionist temporal classification (CTC) have gained substantial research interest. In this work, we extended our previous approach towards training CTC-based systems multilingually. Our systems feature a global phone set, based on the joint phone sets of each source language. We evaluated the use of different language combinations as well as the addition of Language Feature Vectors (LFVs). As contrastive experiment, we built systems based on graphemes as well. Systems having a multilingual phone set are known to suffer in performance compared to their monolingual counterparts. With our proposed approach, we could reduce the gap between these mono- and multilingual setups, using either graphemes or phonemes.

Matthias Sperber, Graham Neubig, Jan Niehues et al.

We investigate the problem of manually correcting errors from an automatic speech transcript in a cost-sensitive fashion. This is done by specifying a fixed time budget, and then automatically choosing location and size of segments for correction such that the number of corrected errors is maximized. The core components, as suggested by previous research [1], are a utility model that estimates the number of errors in a particular segment, and a cost model that estimates annotation effort for the segment. In this work we propose a dynamic updating framework that allows for the training of cost models during the ongoing transcription process. This removes the need for transcriber enrollment prior to the actual transcription, and improves correction efficiency by allowing highly transcriber-adaptive cost modeling. We first confirm and analyze the improvements afforded by this method in a simulated study. We then conduct a realistic user study, observing efficiency improvements of 15% relative on average, and 42% for the participants who deviated most strongly from our initial, transcriber-agnostic cost model. Moreover, we find that our updating framework can capture dynamically changing factors, such as transcriber fatigue and topic familiarity, which we observe to have a large influence on the transcriber's working behavior.

Thomas Zenkel, Ramon Sanabria, Florian Metze et al.

Connectionist Temporal Classification has recently attracted a lot of interest as it offers an elegant approach to building acoustic models (AMs) for speech recognition. The CTC loss function maps an input sequence of observable feature vectors to an output sequence of symbols. Output symbols are conditionally independent of each other under CTC loss, so a language model (LM) can be incorporated conveniently during decoding, retaining the traditional separation of acoustic and linguistic components in ASR. For fixed vocabularies, Weighted Finite State Transducers provide a strong baseline for efficient integration of CTC AMs with n-gram LMs. Character-based neural LMs provide a straight forward solution for open vocabulary speech recognition and all-neural models, and can be decoded with beam search. Finally, sequence-to-sequence models can be used to translate a sequence of individual sounds into a word string. We compare the performance of these three approaches, and analyze their error patterns, which provides insightful guidance for future research and development in this important area.

Jan Niehues, Eunah Cho, Thanh-Le Ha et al.

In this paper, we offer an in-depth analysis about the modeling and search performance. We address the question if a more complex search algorithm is necessary. Furthermore, we investigate the question if more complex models which might only be applicable during rescoring are promising. By separating the search space and the modeling using $n$-best list reranking, we analyze the influence of both parts of an NMT system independently. By comparing differently performing NMT systems, we show that the better translation is already in the search space of the translation systems with less performance. This results indicate that the current search algorithms are sufficient for the NMT systems. Furthermore, we could show that even a relatively small $n$-best list of $50$ hypotheses already contain notably better translations.

Robin Ruede, Markus Müller, Sebastian Stüker et al.

Using supporting backchannel (BC) cues can make human-computer interaction more social. BCs provide a feedback from the listener to the speaker indicating to the speaker that he is still listened to. BCs can be expressed in different ways, depending on the modality of the interaction, for example as gestures or acoustic cues. In this work, we only considered acoustic cues. We are proposing an approach towards detecting BC opportunities based on acoustic input features like power and pitch. While other works in the field rely on the use of a hand-written rule set or specialized features, we made use of artificial neural networks. They are capable of deriving higher order features from input features themselves. In our setup, we first used a fully connected feed-forward network to establish an updated baseline in comparison to our previously proposed setup. We also extended this setup by the use of Long Short-Term Memory (LSTM) networks which have shown to outperform feed-forward based setups on various tasks. Our best system achieved an F1-Score of 0.37 using power and pitch features. Adding linguistic information using word2vec, the score increased to 0.39.

Matthias Sperber, Graham Neubig, Jan Niehues et al.

The input to a neural sequence-to-sequence model is often determined by an up-stream system, e.g. a word segmenter, part of speech tagger, or speech recognizer. These up-stream models are potentially error-prone. Representing inputs through word lattices allows making this uncertainty explicit by capturing alternative sequences and their posterior probabilities in a compact form. In this work, we extend the TreeLSTM (Tai et al., 2015) into a LatticeLSTM that is able to consume word lattices, and can be used as encoder in an attentional encoder-decoder model. We integrate lattice posterior scores into this architecture by extending the TreeLSTM's child-sum and forget gates and introducing a bias term into the attention mechanism. We experiment with speech translation lattices and report consistent improvements over baselines that translate either the 1-best hypothesis or the lattice without posterior scores.

Jan Niehues, Eunah Cho, Thanh-Le Ha et al.

Recently, the development of neural machine translation (NMT) has significantly improved the translation quality of automatic machine translation. While most sentences are more accurate and fluent than translations by statistical machine translation (SMT)-based systems, in some cases, the NMT system produces translations that have a completely different meaning. This is especially the case when rare words occur. When using statistical machine translation, it has already been shown that significant gains can be achieved by simplifying the input in a preprocessing step. A commonly used example is the pre-reordering approach. In this work, we used phrase-based machine translation to pre-translate the input into the target language. Then a neural machine translation system generates the final hypothesis using the pre-translation. Thereby, we use either only the output of the phrase-based machine translation (PBMT) system or a combination of the PBMT output and the source sentence. We evaluate the technique on the English to German translation task. Using this approach we are able to outperform the PBMT system as well as the baseline neural MT system by up to 2 BLEU points. We analyzed the influence of the quality of the initial system on the final result.

Thanh-Le Ha, Jan Niehues, Alex Waibel

In this paper we combine the advantages of a model using global source sentence contexts, the Discriminative Word Lexicon, and neural networks. By using deep neural networks instead of the linear maximum entropy model in the Discriminative Word Lexicon models, we are able to leverage dependencies between different source words due to the non-linearity. Furthermore, the models for different target words can share parameters and therefore data sparsity problems are effectively reduced. By using this approach in a state-of-the-art translation system, we can improve the performance by up to 0.5 BLEU points for three different language pairs on the TED translation task.