Presentation at the Association for Machine Translation in the Americas (AMTA) in Boston, March 2018.
This presentation gives the audience an overview:
- How chatbots are built,
- How they are being used to accelerate and scale the operations of Support and Customer Success teams
- How to create a multilingual chatbot
- How MT can be used to turn a monolingual chatbot into a multilingual agent
2. What is a
chatbot
A chatbot is a computer program designed to
interact with users through a messaging (chat)
service in a way that is designed to seem like a
conversation.
It provides a friendlier, more responsive way to
interact with people by letting them communicate
more naturally and without delays. In some
cases, in addition to answering the chatbot’s
questions, users can also ask the chatbot simple
questions.
NLP and AI are usually employed to achieve a
genuine conversational UX. The application can
be leveraged to both provide and gather data
from users.
4. Chatbots’
reason to be
User Experience
• Convenient access to relevant data in any
situation
• Messaging apps: Over 4 billion users
• Alleviate frustration caused by traditional
means and agents
• Always on, no wait time (and easier to
hang up on them!)
Strategic Advantage
• Reduced manual labor
• Scalability and long-term savings
• Avid data collection
5. A case for (multilingual) chatbots
• 90% of businesses use Facebook to respond to service
requests
• Customers are 5 times more likely to message a company
than posting on its Facebook page
• The average messaging conversation is 66% longer than the
average page conversation
• 10 hours wait: Average time it takes for a company to respond
to a message (about half if you post on their public wall)
• 56% of businesses say engagement through messaging is ROI
positive; 58% say it reduces costs
Source: “Data: A Massive, Hidden Shift Is Driving Companies To Use A.I. Bots Inside Facebook
Messenger,” BusinessInsider.com, May 12, 2016
6.
7.
8. Types of chatbots
Flow type
• Diagram-driven
(digital switchboard)
• Highly specialized
• Highly structured
• Single data source
1
AI bots
• Predictive
• NLP + Machine
Learning
• (Big) Data-driven
• Multiple data sources
2
Hybrids
• Flow type + NLP
• Human-assisted
3
11. Chatbot design:
Best practices
• Keep communication simple
• Short messages
• Simple grammar
• Don’t pretend to be a human
• Avoid witty language
• Sarcasm
• Double meanings
• Humor
• If your first language is not good enough,
don’t add any more yet
12. Content design
for MT:
Best practices
• Keep communication simple
• Short messages
• Simple grammar
• Don’t pretend to be a human
• Avoid witty language
• Sarcasm
• Double meanings
• Humor
• If your first language is not good enough,
don’t add any more yet
13. All the usual L10N challenges apply
Ambiguity Context Gender Culture
Lingo, humor,
sarcasm
Tone Intent
Source
quality
14. MT Opportunities:
Multilingual AI bots
• Help AI chatbot developers translate
and refine data models
• Intents
• Entities
• Phrase tables
• Translate integrated data sources
• Turn discarded utterances into data
(intent, entities…)
• Potential as continuous revenue
stream
16. MT opportunities:
Flow type bots
• Translate all entities and linked
text/articles
• Leverage existing enterprise
language resources
• Rapid implementation
• Most interesting add-value service
17. MT opportunities:
Hybrid models
• Most common type
• Error tolerance
• Larger data sources
• Knowledge bases
• Product catalogs
• Ticketing systems
• Reward systems