In this guest post, Director of Strategic Initiatives at ISI/USC Daniel Marcu shares the findings of his survey carried out in support of the 2016 AMTA MT Commercialization Panel, as well as his own opinion on the present and future of machine translation.
On October 31st 2016, I moderated an MT Commercialization Panel at the 2016 Conference of the Association for Machine Translation in the Americas. The panel participants were Macduff Hughes, Engineering Director of Google Translate; Valery Jacot, Software Development Manager, Localization Platform, Autodesk; Dragos Munteanu, Director of Research and Development MT, SDL plc.; and Chris Wendt, Group Manager, Machine Translation, Microsoft. In support of the panel, I carried out a survey that drew participants of various backgrounds:
- 66 (34.2%) Machine Translation Researchers or Practitioners;
- 55 (28.5%) Professional Translators or Project Managers;
- 30 (15.5%) Language Technology Executives;
- 12 (6.22%) Language Services Executives;
- 30 (15.5%) Other professionals associated with the industry (Consultants, System Integrators, Research Analysts, Vendor Managers, Solution Managers, Funders, etc).
In this post, I summarize my findings. For each of the eight questions in the survey, I present the results in the aggregate, across all five backgrounds; and the results specific to the four largest cohesive categories: Machine Translation Researchers or Practitioners, Professional Translators or Project Managers, Language Technology Executives, and Language Services Executives. For each question, I make explicit the set of answers participants could select from and summary statistics in the aggregate and for the four categories.
Q1: What are the top barriers to broader machine translation adoption in the Language Services industry (check all that apply)?
- A1: Machine Translation Quality (too low to be useful)
- A2: Machine Translation Inconsistency (sometimes it is good, sometimes it is bad)
- A3: Risk: LSPs and Translators don't know how to price MT jobs correctly
- A4: Poor integration into existing CAT tool
- A5: Most professional translators do not know how to use MT and be productive
- A6: Most professional translators do not want to use MT. Their job is no longer fun when they try to do it
- A7: Most professional translators are paid less when they post-edit MT and they make less money overall
- A8: Most customers require LSPs to not use MT
- A9: Other
In the aggregate, MT Inconsistency (A1) and MT Quality (A2) represent the top choices with pressure exercised by clients to not use MT ranking last (A8). However, there are significant differences in perspective across the four groups of participants in the survey. Language Technology Executives and MT Researchers and Practitioners believe that professional translators do not know how to use MT (A5); while Language Services Executives and Translators are concerned that when using MT their revenues or compensation goes down (A7). Language Services Executives are also worried that they do not know how to price MT post-editing jobs correctly (A3), while Translators and Project Managers are the group that ranks highest in acknowledging that some translators buyers ask translators explicitly to not use MT (A8); this is a concern that is not on the radar of MT researchers and practitioners.
Q2: Increased machine translation quality will have the following impact on the Language Services market
- A1: The Language Services market will grow. Clients will want to translate and post-edit more. Only a fraction of the translation needs are met today.
- A2: The Language Services market will shrink. Better MT will lead to lower services costs.
- A3: The Language Services market will not change. More stuff will get translated overall, but at lower translation costs.
The vast majority of the participants believe that improvements in MT quality will increase the Language Services market (51%) or will have no impact on it (33%). The categories that are most worried about increases in translation quality because they perceive them as leading to the Language Services shrinking are the groups doing the work: Translators and Project Managers (27%) and Language Services Executives (25%).
Q3: Increased machine translation quality will have the following impact on the professional translators
- A1: Professional translators will make more money. They will be able to get more stuff done.
- A2: Professional translators will make less money. They will be squeezed into accepting increasingly lower translation/post-editing prices.
- A3: Technology savvy professional translators will strive. The others will struggle.
- A4: Increased MT quality will have no impact on professional translators
In the aggregate, participants believe that technology savvy professionals will eventually replace those who are not (A3); only 20% of the survey participants believe that increased MT quality will lead to translators making less money. The only category that is pessimistic is the translators and project managers category: only 38% believe that being technology savvy will help as translation gets better while 45% believe that it is likely that their compensation will go down.
Q4: Increased machine translation quality will have the following impact on MT researchers
- A1: The number of MT researchers will grow. More companies will want to own MT capabilities.
- A2: The number of MT researchers will shrink. Once the top providers offer high-quality MT, everyone will buy that service from them.
- A3: There will be no change in the number of researchers working in MT.
This is the question that elicited the highest agreement among survey participants. Most participants believe that the number of MT researchers will grow.
Q5: Increased machine translation quality will have the following impact on the Language Services Industry
- A1: The industry will begin to consolidate. Technology savvy LSPs will steal business from those who are not.
- A2: The industry will fragment further. It will become easier to create LSP businesses and be successful.
- A3: There will be no changes in the industry. MT quality is not an enabler of market consolidation or a driver of market fragmentation.
This is another question that was characterized by a surprising degree of consistency in the answers. Most participants (54%) believe that increased MT quality will lead to the industry being consolidated; 25% believe that the industry will not change; and only 18% believe that improved MT quality will lead to further fragmentation.
Q6: The top 10 Language Services companies have combined revenues that represent less than 10% of the market. Why is the Language Services market so fragmented? (check all that apply)
- A1: The industry has a natural, long tail of buyers that is difficult to reach
- A2: Most enterprises do not have a good handle on their translation processes and costs. This fragmentation inhibits market consolidation and fast growth
- A3: Buyer-vendor relationships in the translation industry are built on trust and are difficult to disrupt
- A4: Language Services companies cannot grow easily beyond $100M in revenues. They do not provide sufficient differentiation to steal business away from other competitors.
- A5: No credible high-quality, low-cost, fast turn-around translation provider has emerged to date
- A6: The talent pool required to meet the needs of the industry is limited and loyal to the service providers they work with
This is a question that elicited a variety of opinions: 47% of the participants attribute the fragmentation of the industry to the difficulty of reaching the long tail of buyers (A1) and to the fact that most enterprises do not have a good handle of their translation processes - they do not know how much stuff they translate and how much that costs them (A2). However, significant differences exist across the four categories: Language Technology executives feel that it is often difficult to disrupt long-term, trust-based relationships between enterprises and vendors; Language Services executives feel that differentiation is difficult to establish, from the outside, most LSPs looking the same; Translators and Project Managers assign, more or less, the same percentage to all categories and are the ones who attribute the highest importance to the limited, loyal talent pool they represent. Interestingly, although they perceive themselves as being loyal to the service providers they work with (38%), the feeling is not mutual: Language Services executives believe that the loyalty of the talent pool cannot be counted on (only 8% voted for this alternative).
Q7: What problem should a team solve if it started a machine translation company today?
- A1: Develop an MT engine that provides higher quality than any other commercial MT engine
- A2: Develop MT-related technology that helps Language Service providers
- A3: Develop MT-related technology that helps professional translators
- A4: Develop high-value, end-to-end commercial solution that tolerates less than perfect MT
- A5: Don't start an MT company. Too late to do that.
In the aggregate, the opinions range across all five answers. However, Language Technology and Services Executives feel strongly that MT startups should focus on developing end-to-end solutions that tolerate less than perfect MT (47% and 58% respectively). Only Translators and Project Managers believe that focusing on solutions for their group/market is a good idea (45%). Surprisingly, 89% of the participants believe it is still a good idea to start an MT company.
Q8: How many machine translation providers will we have in 10 years?
- A1: Less than 5
- A2: More than 5 less than 20
- A3: More than 20
The majority (51%) of the participants believe that 10 years from now we will have between 5 and 20 providers; 35% believe that we will have more than 20 providers; and only 11% believe that we will have less than 5. Most participant categories follow a similar distribution. All in all, by carrying this survey, I was most surprised by the differences in opinion across the four groups of participants. If you are a player in this market, keep that in mind when interacting with other parties in the industry.
How would I answer the questions in my survey?
- Q1: I believe that almost all suggested answers associated with this question represent valid barriers to adoption. Some more important than others; and a few more. I checked boxes A1, A2, A3, A5, and A6.
- Q2: During the last ten years, the Language Services Market has been growing steadily in spite of significant improvements in MT quality and increased MT adoption. I do not see any good reason for this growth to stop during the next five to ten years.
- Q3: Machine Translation will not replace professional translators; but technology savvy professional translators will replace those who do not embrace and adapt to change.
- Q4: The number of MT researchers will grow in the industry and shrink in the academia; commercial enterprises will invest more into MT R&D than government sponsoring agencies.
- Q5: MT quality alone is not a driver of consolidation in the Language Services industry. Several other technology and business innovations are required to consolidate this industry.
- Q6: Many of the answers associated with this question explain in various degrees the fragmentation of the Language Services industry. But to me, the most important one is that no trusted, high-quality, low-cost, fast turn-around language service provider has emerged to date.
- Q7: The time to start an MT company has passed. Nobody should do it unless they are happy to be acquired for a modest sum after lots of frustration and hard work; or go bankrupt. AMTA
- Q8: We will have only a handful of MT providers 10 years from now. The investment required to create and sustain a competitive MT service can be afforded only by a handful of global technology giants for whom MT is an important business asset.
About the author
Daniel is recognized as a leading authority in natural language processing and successful entrepreneur. He has co-authored an MIT Press book, more than 100 peer-reviewed articles and 30 USPTO patents. Daniel is the Director of Strategic Initiatives at ISI/USC and the founder of Symbyonics, LLC. You can find his LinkedIn profile here.