ALGORITHMIC TEXTUAL PRACTICES: IMPROVING FLUENCY AND WORD ORDER IN NEURAL MACHINE TRANSLATION OUTPUT.
Neural machine translation constitutes a substitute to the statistical machine translation pattern in diverse languages. (Eriguchi et al., 2019) The prevailing neural machine translation patterns commonly employ word-level modeling to insert input sentences into semantic space (Balica, 2018; Cruciani, 2018; Fabricio, 2016; Lazaroiu, 2018; Mihaila and Mateescu, 2017; Nica, 2018; Popescu, 2017a, b; Radulescu, 2017), but that may not be adequate for the encoder modeling of neural machine translation, particularly for languages where tokenizations are typically confusing, as there may be tokenization inaccuracies which may adversely shape the encoder modeling of neural machine translation, while the adequate tokenization granularity is ambiguous. (Tan et al., 2018)
2. Conceptual Framework and Literature Review
Neural machine translation represents a groundbreaking model, and the attention mechanism (Bolton et al., 2018; Dicu, 2018; Farber, 2017; Machan, 2017; Mihaila, 2018; Nica et al., 2018; Popescu et al., 2017; Sion, 2018a, b) is the prominent method with the cutting-edge records in numerous language pairs. All options of the attention mechanism employ only temporal awareness where one scalar value is allocated to one context vector matching a source word. (Choi et al., 2018) Undeveloped neural machine translation patterns are contingent on sequence-to-sequence learning that expresses in code a sequence of source words into a vector space and produces an additional sequence of target words from the vector: sentences are interpreted just as sequences of words with no internal structure. (Eriguchi et al., 2019) The naive back-translation method enhances the translation performance considerably, but its application for monolingual input is not relevantly valid as established neural machine translation patterns do not differentiate between the genuine parallel corpus and the back translated synthetic parallel one. (Yang et al., 2019)
3. Methodology and Empirical Analysis Building my argument by drawing on data collected from the Boston Consulting Group, Deloitte, eMarketer, Locaria, MIT Sloan Management Review, NCSC, and Statista, I performed analyses and made estimates regarding awareness and usage of translation applications featuring machine learning (%), how professional translators worldwide see artificial intelligence affecting their work in the future (%), and market size of the global language services industry (billion U.S. dollars). The results of a study based on data gathered from 4,200 respondents provide support for my research model. Employing the structural equation modeling and using the probability sampling technique, I collected and inspected data via a self-administrated questionnaire.
4. Results and Discussion
Current developments of neural machine translation are instrumental in performing translation by employing an uncomplicated end-to-end architecture. (Eriguchi et al., 2019) Similarly as any machine learning algorithms, neural networks represent a definite class of statistical learning patterns. (Petrucci et al., 2018) Machine translation is presently going through a paradigm repositioning from statistical to neural network patterns. (Moorkens, 2018) Neural machine translation generates significantly fewer lexical, morphology, and word order errors, better configurations in the rearrangement of verbs and nouns, but produces more inaccuracies in the translation of proper nouns than phrase-based approaches. (Bentivogli et al., 2018) Nearly all applications of neural machine translation patterns use a padding approach when processing a mini-batch to alter to the same length all sentences in it. (Qiao et al., 2018) Statistical machine translation systems are disadvantaged by fluency errors, notably as grammatical inaccuracies and ones concerning idiomatic word choices. (Tezcan et al., 2019) (Tables 1-6)
5. Conclusions and Implications
Neural machine translation is ordinarily enhanced to produce sentences which comprise n-grams with ground target to the fullest extent, but n-gram precisions, the physically fashioned approximate loss function, can misdirect the pattern to create suboptimal translations. (Yang et al., 2018) Handling the ontology learning problem as a neural machine translation assignment may be a sound manner to approach perpetual expressive ontology learning tasks (Bratu, 2018; Donnelly, 2017; King, 2017; Michailidou, 2018; Nica, 2017; Peters, 2017; Popescu, 2018; Skordoulis, 2016; Smith and Stirling, 2018), such as language inconsistency, domain self-determination, and significant engineering expenses. (Petrucci et al., 2018) Neural networks integrate frame of reference from the training input, source text, and elaborating target text and mostly generate words in the adequate context, although systems are nevertheless inconstant and necessitate human review. (Taivalkoski-Shilov, 2018)
This paper was supported by Grant GE-1287672 from the Cognitive Labor Institute, New York, NY.
The author confirms being the sole contributor of this work and approved it for publication.
Conflict of Interest Statement
The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Balica, R. (2018). "Big Data Learning Analytics and Algorithmic Decision-Making in Digital Education Governance," Analysis and Metaphysics 17: 128-133.
Bentivogli, L., A. Bisazza, M. Cettolo, and M. Federico (2018). "Neural versus Phrasebased MT Quality: An In-Depth Analysis on English-German and English-French," Computer Speech & Language 49: 52-70.
Bolton, C, V. Machova, M. Kovacova, and K. Valaskova (2018). "The Power of Human-Machine Collaboration: Artificial Intelligence, Business Automation, and the Smart Economy," Economics, Management, and Financial Markets 13(4): 51-56.
Bratu, S. (2018). "The Ethics of Algorithmic Sociality, Big Data Analytics, and Data-Driven Research Patterns," Review of Contemporary Philosophy 17: 100-106.
Choi, H., K. Cho, and Y. Bengio (2018). "Fine-Grained Attention Mechanism for Neural Machine Translation," Neurocomputing 284: 171-176.
Cruciani, M. (2018). "Explicit Communication: An Interest and Belief-based Model," Linguistic and Philosophical Investigations 17: 50-70.
Dicu, G. (2018). "Linguistic Practices in Gendered Institutional Settings: Speech Behavior of Female Political Leaders," Journal of Research in Gender Studies 8(1): 158-164.
Donnelly, R. (2017). "Blended Problem-based Learning in Higher Education: The Intersection of Social Learning and Technology," Psychosociological Issues in Human Resource Management 5(2): 25-50.
Eriguchi, A., K. Hashimoto, and Y. Tsuruoka (2019). "Incorporating Source-Side Phrase Structures into Neural Machine Translation," Computational Linguistics. doi:10.1162/COLI_a_00348.
Fabricio, B. F. (2016). "Repetition-Differentiation in an Online Debate on Sexualities: Re-Imagining Identity Meanings," Journal of Research in Gender Studies 6(1): 241-261.
Farber, M. (2017). "The Future of Journalistic Labor in the Age of Digital Narratives: The Algorithmic Authority of Automated News as a Legitimate Knowledge Producer," Psychosociological Issues in Human Resource Management 5(2): 199-204.
King, D. (2017). "Indeterminacy of Translation and Innate Concepts: A Critical Review," Linguistic and Philosophical Investigations 16: 104-141.
Lazaroiu, G. (2018). "Postmodernism as an Epistemological Phenomenon," Educational Philosophy and Theory 50(14): 1389-1390.
Machan, T. R. (2017). "Can People Obtain Objective Knowledge?," Psychosociological Issues in Human Resource Management 5(1): 90-119.
Michailidou, A. (2018). "Parallel Narratives, Colorful Visions: Enclosure and Liberation in Tracy Chevalier's The Virgin Blue," Journal of Research in Gender Studies 8(2): 11-28.
Mihaila, R., and M. Mateescu (2017). "The Gender Politics of Postfeminist Semantics," Journal of Research in Gender Studies 7(1): 255-261.
Mihaila, R. (2018). "The Ascendance of Postmodernism in the Educational Sphere," Educational Philosophy and Theory 50(14): 1578-1579.
Moorkens, J. (2018). "What to Expect from Neural Machine Translation: A Practical In-Class Translation Evaluation Exercise," The Interpreter and Translator Trainer 12(4): 375-387.
Nica, E. (2017). "Techno-Pedagogy Knowledge in Smart Learning Environments," Economics, Management, and Financial Markets 12(1): 75-81.
Nica, E. (2018). "The Social Concretisation of Educational Postmodernism," Educational Philosophy and Theory 50(14): 1659-1660.
Nica, E., C. Manole, and C. I. Stan (2018). "A Laborless Society? How Highly Automated Environments and Breakthroughs in Artificial Intelligence Bring About Innovative Kinds of Skills and Employment Disruptions, Altering the Nature of Business Process and Affecting the Path of Economic Growth," Journal of Self-Governance and Management Economics 6(4): 25-30.
Peters, M. A. (2017). "Disciplinary Technologies and the School in the Epoch of Digital Reason: Revisiting Discipline and Punish after 40 Years," Contemporary Readings in Law and Social Justice 9(1): 28-46.
Petrucci, G., M. Rospocher, and C. Ghidini (2018). "Expressive Ontology Learning as Neural Machine Translation," Journal of Web Semantics 52/53: 66-82.
Popescu, G. H. (2017a). "The Economics of E-learning Production," Economics, Management and Financial Markets 12(1): 68-74.
Popescu, G. H. (2017b). "Organizational E-learning and Knowledge Management in Higher Education," Journal of Self-Governance and Management Economics 5(1): 87-93.
Popescu, G. H., M. Comanescu, and C. Manole (2017). "Mobile Knowledge Work, Information Routines, and Digital Technologies," Psychosociological Issues in Human Resource Management 5(2): 187-192.
Popescu, G. H. (2018). "Has Postmodernism the Potential to Reshape Educational Research and Practice?," Educational Philosophy and Theory 50(14): 1490-1491.
Qiao, Y., K. Hashimoto, A. Eriguchi, H. Wang, D. Wang, Y. Tsuruoka, et al. (2018). "Parallelizing and Optimizing Neural Encoder-Decoder Models without Padding on Multi-Core Architecture," Future Generation Computer Systems. doi:10.1016/j.future.2018.04.070.
Radulescu, A. (2017). "The Underlying Cognitive Mechanisms Involved in Consecutive Interpreting," Linguistic and Philosophical Investigations 16: 142-148.
Sion, G. (2018a). "Smart Educational Ecosystems: Cognitive Engagement and Machine Intelligence Algorithms in Technology-Supported Learning Environments," Analysis and Metaphysics 17: 140-145.
Sion, G. (2018b). "How Artificial Intelligence Is Transforming the Economy. Will Cognitively Enhanced Machines Decrease and Eliminate Tasks from Human Workers through Automation?," Journal of Self-Governance and Management Economics 6(4): 31-36.
Skordoulis, K. (2016). "Science, Knowledge Production and Social Practice," Knowledge Cultures 4(6): 291-307.
Smith, A., and A. Stirling (2018). "Innovation, Sustainability and Democracy: An Analysis of Grassroots Contributions," Journal of Self-Governance and Management Economics 6(1): 64-97.
Taivalkoski-Shilov, K. (2018). "Ethical Issues Regarding Machine(-assisted) Translation of Literary Texts," Perspectives. doi: 10.1080/0907676X.2018.1520907
Tan, Z., J. Su, B. Wang, Y. Chen, and X. Shi (2018). "Lattice-to-Sequence Attentional Neural Machine Translation Models," Neurocomputing 284: 138-147.
Tezcan, A., V. Hoste, and L. Macken (2019). "Estimating Word-Level Quality of Statistical Machine Translation Output Using Monolingual Information Alone," Natural Language Engineering. doi:10.1017/S1351324919000111
Yang, Z., W. Chen, F. Wang, and B. Xu (2018). "Generative Adversarial Training for Neural Machine Translation," Neurocomputing 321: 146-155.
Yang, Z., W. Chen, F. Wang, and B. Xu (2019). "Effectively Training Neural Machine Translation Models with Monolingual Data," Neurocomputing 333: 240-247.
Spiru Haret University, Bucharest, Romania
How to cite: Radulescu, Adina (2019). "Algorithmic Textual Practices: Improving Fluency and Word Order in Neural Machine Translation Output," Linguistic and Philosophical Investigations 18: 126-132.
Received 17 December 2018 * Received in revised form 18 March 2019
Accepted 19 March 2019 * Available online 14 April 2019
Table 1 Awareness and usage of smartphone applications featuring machine learning (%) Awareness Usage Predictive text 54 38 Route suggestions 40 26 Voice assistants 39 15 Voice search 35 14 Translation apps 31 16 Voice-to-text 27 12 Email classification 26 11 Automated calendar entries 22 9 Location-based app suggestions 21 11 Automated photo classification 21 8 Sources: Deloitte; Statista; my survey among 4,200 individuals conducted November 2018. Table 2 Market size of the global language services industry (billion U.S. dollars) 2009 23.5 2010 26.3 2011 28.34 2012 33.05 2013 34.78 2014 37.19 2015 38.16 2016 40.27 2017 43.08 2018 46.52 2019 49.14 Sources: Statista; my estimates. Table 3 Do you utilize remote interpreting technology? (Check all that apply, %) Audio or telephonic interpretation (standard telephone) 88 Video conferencing 61 Video remote interpreting 52 Web-based applications (e.g. Skype) 48 Specialized telephone equipment 34 Other technological solutions (please provide details) 23 Voice over internet protocol 21 Translation software and automated interpreter software 18 N/A 7 Sources: NCSC; my survey among 4,200 individuals conducted November 2018. Table 4 Levels of understanding for artificial intelligence-related technology and business context: To what extent do you agree with the following statements about your organization? (%) We understand... Pioneers Investigators Technology Required 92 78 implications technological breakthroughs to succeed with artificial intelligence Technology Data required for 87 76 implications artificial intelligence algorithm training Technology Processes for 86 70 implications artificial intelligence algorithm training Business Artificial 92 84 implications intelligence- related changed ways of business value generation Business Development 85 74 implications time of artificial intelligence- based products and services Business Development 83 75 implications costs of artificial intelligence- based products and services Workplace Required changes 84 84 implications of knowledge and skills for future artificial intelligence needs Workplace Effect of artificial 87 78 implications intelligence in the workplace on organization's behavior Industry Artificial 83 83 implications intelligence- related shift of industry power dynamics Experimenters Passives Technology 27 15 implications Technology 24 11 implications Technology 20 7 implications Business 33 22 implications Business 21 18 implications Business 16 10 implications Workplace 22 19 implications Workplace 18 18 implications Industry 22 18 implications Sources: MIT Sloan Management Review; The Boston Consulting Group; my survey among 4,200 individuals conducted November 2018. Table 5 How professional translators worldwide see artificial intelligence affecting their work in the future (%) No impact whatsoever 4 Less work for professional translators 46 Same volume but higher standards expected 37 Overall volume of work will grow 13 Sources: MIT Sloan Management Review; The Boston Consulting Group; my survey among 4,200 individuals conducted November 2018.
|Printer friendly Cite/link Email Feedback|
|Publication:||Linguistic and Philosophical Investigations|
|Date:||Jan 1, 2019|
|Previous Article:||ARE THE WORKERS IN THE ON-DEMAND ECONOMY EMPLOYEES, INDEPENDENT CONTRACTORS, OR A HYBRID CATEGORY?|
|Next Article:||SELF-PORTRAITS IN SOCIAL MEDIA: MEANS OF COMMUNICATING EMOTION THROUGH VISUAL CONTENT-SHARING APPLICATIONS.|