This chatbot is a proof-of-concept for a story-telling AI.
A story is generated with several characters, locations and objects. The characters interact with one another, can pick up objects and move from a location to another one. The user will be able to ask the chatbot questions about the story and even make suggestions to the chatbot in order to influence the story.
This is a fourth-year computer engineering project which aims to teach us about Python and natural-language processing. It is a widely studied computer science field, allowing many real-world applications such as personal assistants and real-time translation.
Natural-language processing has many challenges we encountered. Because of the small scale of this project, we didn't attempt to solve them: for example, syntaxic and semantic ambiguity is one of the most apparent flaw of the chatbot. We believe nonetheless that NLP is a major field we will take time to discover.
The chatbot is able to answer using three modes. Mode 3 is the main story mode. It uses a system of tags: the user's input sentece is associated with a list of tags generated from a database of about 60 tags. This list is used to find a suitable answer in another database which contains several answers, each one corresponding to a tag list.
Mode 2 is a simplified Mode 3 : in case Mode 3 is not able to compute an answer, Mode 2 will give a simpler answer by re-associating the same sentence with only one tag.
Mode 1 does not use tag recognition and whatever the user says, the chatbot will answer with 'backchannels', such as Uhh.. or Hmm….
Further explanations can be found in French (/misc/Rapport.pdf).
The NLTK library was used, especially its tokenize function: it is able to separate contractions and negations. Our chatbot is thus able to recognize I'm and I am as the same expression, for example.
To run the program, several modules are to be used: NLTK, NumPy and Re. For NLTK, downloading an extra package may be required:
pip install nltk
python -m nltk.downloader 'punkt'
The punkt package is included as well with this git (/nltk_data/tokenizers). In case the NLTK downloader is broken, you can copy the whole folder into:
- Windows:
C:\nltk_data\tokenizers - Mac OS:
/usr/local/share/nltk_data/tokenizers - Unix:
/usr/share/nltk_data/tokenizers
The command used to run the project is:
python main.py
When running the chatbot, the user can ask questions such as:
- Where is Annie?
- How is she doing?
To which the chatbot will be able to answer:
- Annie is in the kitchen.
- Annie is feeling depressed.
Make sure your Python 3 install is up to date as some functions have been renamed (iteritems() and .raw_input() for example)
Possible areas for improvment would be:
- adding the possibility to recognize multiple words for one tag (e.g. "Laura's bedroom" is currently recognized as Laura, 's and bedroom)
- adding verbs and noun tagging,
- adding partial answers the chatbot will combine,
- fixing typos such as blank spaces before question marks,
- implementing a 'context' system for the chatbot to remember about the conversation's subject for example,
- adding question recognition without question marks. Our tagging system relies on the ? at the end of each question,
- and of course adding more tags and improving the simulation, with true storytelling and a coherent narrator.
5 CHARACTERS
Annie, Antoine, Maria, Jeffery, Laura.
Each character is defined by their name and last name, their gender, in which room they are in, the objects they carry, their state of mind at the moment, their physical health, their nutrition, hydration and sleep levels, their relationship with the other characters.
THE ROOM(s) :
Living room, kitchen, toilets, bathroom, roof, a bedroom for each character, and the outside.
Each room is defined by : its name, its state (messy, wrecked, tidy…), if people can acces it, who is inside, what it contains (tools, ressources, and entertainments)
THE RESSOURCES :
They are consumed when used, and have different effect depending on what they are. The different ressources that appear in the simulation are: Water, Basic Food, Well Prepared food, Medicines, Alcohol
THE ENTERTAINMENTS :
The different entertainments that appear in the simulation are: TV, Deck of cards, Books, Computer
THE TOOLS :
The different tools available in the simulation are: Crowbar, Gun, Rifle, Torchlamp...
Adrien Lavillonnière (Veados)
Corentin Manscour (neofoetus)
Hien Minh Nguyen (minh-n)
Polytech-Paris Sud