Skip to main content

Cost-ORiented Agile Localisation

Periodic Reporting for period 1 - CORAL (Cost-ORiented Agile Localisation)

Reporting period: 2015-07-01 to 2015-12-31

E-commerce businesses and SMEs have realised that globalisation is essential for their commercial growth and key for being resilient against crises in local markets. When an e-commerce decides to go global, translation is one of the most critical assets. In fact, recent studies show that about half of the consumers would never buy at websites where the information is not available in their language. However, SMEs face important economical and technological barriers that prevent their websites to be appropriately translated into the languages of their potential consumers, leaving many business out of the global market.
This is actually a big loss, given the dimensions of the e-commerce market ($1,552 bn in 2013) and its growth rate (up to 33% from 2010 to 2015 in some regions).
Currently e-commerces attempting to solve this problem can opt for:
Professional or crowdsourced translation services that are perceived as too expensive
Inexpensive machine translation (MT) services whose quality is not enough to achieve consumer trust
Project CORAL will provide an innovative cost-oriented translation service that will allow e-commerces to adopt a lean strategy when entering a new market, by providing a fine-grained, customisable compromise between quality and cost. This will be achieved by using proprietary technology that:
Identifies the most relevant texts, at the page level, that should be assigned to human translators
Lowers the cost for less important texts by using specialised MT
The goal of the feasibility study conducted during Phase I was to validate the main hypothesis supporting the CORAL project, i.e.:
Value of the technology. CORAL assumes that e-stores need to cut down drastically in translation costs, demanding hence a compromise between cost and quality. Task 1 deals with this hypothesis.
Acceptability of the technology. CORAL relies on the preliminary observation that allocating professional translation to certain strings, and MT to the rest, will yield similar perceived quality as if though the whole text was professionally translated. This hypothesis was tested in Task 2.
Assessment of the revenue streams. CORAL was designed to generate revenue by means of commission on top of sales in the languages offered. The viability of this scheme is analysed in detail in Task 5.
Economic viability of CORAL. An in-depth business plan was developed in collaboration with Valencia’s CEEI (European Centre for Innovative Enterprises) and is reported in Task 6.
The work performed during this reporting period (which is the totality of the project), concerns all six foreseen tasks:

Task1: Product-market analysis
To assess whether CORAL really meets the needs of the e-commerce market, we conducted a study in which several e-commerce owners, potential targets of our technology, were interviewed. Even though the solution offered by CORAL is language-agnostic, we conducted this study by focusing on the Spanish e-commerce market, since we understand that geographical proximity and language offers us an important advantage when contacting such e-commerce owners. These interviews had as consequence changes to the the revenue scheme, the technological core, and the long-term strategy of CORAL.

Task2: A/B testing for assessment of perceived quality.
Under project CORAL, human supervision, which is expensive because of its nature, is assigned to those strings that are more important for generating consumer confidence. This allows us to span a broad range between the cost incurred by when using pure machine translation (typically in the order of 0.2¢/word), and the cost incurred by when using human translators to translate all the texts present in the website (typically in the order of 10¢/word). By selecting those texts that are more important for the quality perceived by potential consumers, our goal is to allow e-commerce stores to establish their own mid-point in that range, hence adapting our service to the translation budget that the specific e-commerce avails of. This combination of machine translated texts and human translated texts, which we have named selective translation, constitutes the main technological hypothesis to be tested in project CORAL, namely, that selective translation and human translation will lead to approximately the same conversion potential. For assessing this, we conducted an A/B test in which a certain product was marketed using different translations, and the conversion results were measured. As a result, all conditions ended up being undistinguishable. Even though surprising, the results obtained in this A/B test present an interesting business opportunity. Given the fact that all three translation types seem to yield similar conversion rates, the assumption under CORAL is even more valid: e-commerces that wish to internationalise their business can choose to start off with a pure machine translation (with perhaps some tweaks in the most visible parts of the website), and then slowly transition to human-translated texts. This will allow e-commerce stores to enter new markets with a lower economic entry barrier, and decide whether to invest more in translations once ROI is being realised. Hence, and even though we did not expect raw machine translation to yield similar conversion rates as professional translation, we understand that the main technical assumption under project CORAL is validated, allowing e-commerce stores a lean approach to the website internationalisation problem.

Task3: Freedom to operate analysis
Since Sciling’s staff doesn’t include legal experts or experts in patent research, we decided to subcontract this task to Nordic Innovators, a research and innovation consultancy agency, after exploring other possible alternatives (such as “Pons Patentes y Firmas”, “SVMMA Asesores”, and “Econet Consultants”), ensuring the best value for money. After Nordic Innovator’s FTO report, and Olleros Abogados more intensive analysis of patent US20140200955, the conclusion is that there is no patent that would prevent Sciling from commercialising the service offered by CORAL, not even in the United States.

Task4: Assessment of the dissemination channels
During the duration of the project grant, the following dissemination channels were studied:
a) Direct mailing to e-store owners
b) Conferences and trade fairs
c) Reports in newsletters
d) Dissemination via channel approach (e-devs and marketplace solutions)
As a result from this analysi
Project CORAL has as objective to uncover a new translation market niche: E-commerce sites that cannot afford professional translation, but at the same time cannot rely on fully automatic MT will now have access to a viable internationalisation service.

In the product-market analysis described in depth in Task 1 of Part 1 of this document, we found a large proportion (roughly, 90%) of e-commerce SMEs that stated explicitly that they did not have the means to translate the contents of their website into the languages of their customers. However, and following CO1, this segment of the e-commerce market is not every e-commerce SME any more, since doing so would render the CORAL project unviable from a business point of view. This, hence, implies that the pay-as-you-go monetization scheme (in which revenue is charged by commission on top of sales in the target languages) needs to be revised, and will need to be studied more in depth when working with the first customers. For instance, a possibility would be to charge a low monthly fee (so as to set up a small economic barrier that ensures that the specific e-store is a business), and then charge commission on top of sales. Nevertheless, this monetization scheme will need to be validated when CORAL is already in the commercialisation stage.

The following numbers show the potential impact of CORAL:
Assuming a 10% penetration into the potential market, 4290 e-commerces would be reached.
An average 100 product-language monthly rotation leads to an average 20k words per month. From this amount, we can assume that only 10% will be translated professionally, i.e. 2k words.
4290 e-commerces with a 2k words/month translation requirement leads to a yearly requirement of 103M words.
Assuming a translator can on average translate 3k words per day, i.e. 720k/year, translating 103M words per year would require a workforce of 143 translators.
peces.png