A Focus on Verbs, Not Nouns
When people listen to stories, they hear about persons, places, and things. When people listen to great stories, they experience events, situations, outcomes and actions. From stories, we learn about the challenges, opportunities, feelings and decisions faced by others in a way that enables us to better understand the entire experience and relate to it.
That is why at OntoInsights, we start our work focused on verbs (focused on what is happening and being experienced). When we start with verbs, we can construct a complex timeline and add in who, what, where, when, and why. This is a powerful approach to storytelling, and a powerful approach to analysis to gain a holistic view of the experience of the storyteller.
This blog post introduces a new starting point for ontology development that is novel in that it focuses on verbs. The approach begins by understanding the events and situations of interest in the domain being described, versus trying to catalog all the possible 'things' in that domain. This, in turn, helps to avoid 'boiling the ocean' caused by never really being able to enumerate all the 'things', and aids in structuring the concepts and their relationships so that they are more understandable and easier to use.
For most people seeking to understand a domain of interest, they start with the things (mostly the 'nouns') in that domain. Indeed, in Stanford's paper, "Ontology Development 101: A Guide to Creating Your First Ontology", the authors say … "Classes are the focus of most ontologies. Classes describe concepts in the domain."
But after years of starting with things (with nouns), work on narratives and linguistics have provided new insights. Perhaps, the better answer is to start with verbs - start with the events and situations in the domain. These are the 'things' that are happening, are being experienced, or should be happening and experienced. And work in linguistics, starting with Davidson in 1967 ("The Logical Form of Action Sentences"), defines an event as a 'thing' with properties and relationships. So, this becomes consistent with the authors' advice in the Stanford paper (even though the paper itself focuses on wines, which are nouns).
What happens when you start with events and situations? The problem space becomes more specific, and the overall number of concepts that need to be defined, or their level of granularity, becomes much smaller. One cannot disagree that there are far fewer verbs than nouns, in English at least. And one's focus changes to the nouns that are necessary and relevant to the events and situations of interest. (For more information and background, listen to the video recording of the Ontology Summit 2021 session from March 17th. These topics are addressed in the segment starting around 12:20pm and running for approximately 20 minutes.)
Aside: Note that talking about verbs versus nouns is a bit of a linguistic nightmare, since any verb can be turned into a noun - e.g., 'the bomb exploded violently' versus 'the violent explosion was caused by a bomb'. So, let's just focus on events and situations - on the things that are and should be happening and experienced. Narrative analysis and language processing also bring you back to events and situations, since stories deal with how people understand and explain their worlds, and thereby structure and utilize their 'knowledge'.
OntoInsights has taken these insights and combined them with over 50 years of modeling and ontology experience. We designed a new top-level ontology - one that is focused on events, situations and the people and things experiencing or involved in them. There is no magic in the approach or deep philosophical insights. There is instead a focus on the '6Ws' (who, what, where, when, why and how):
- Happening/occurrence/state (what)
- Agents/actors/participants (who and what)
- Location (where)
- Time/sequence (when)
- Tools/instruments/resources used or provided/consumed (what and how)
- Goal/intent (why)
- Causation, precondition and prevention (why)
- Modalities (planned/future, possible/impossible, attempted, …)
- Other relationships such as collections/groupings, component parts, etc.
To review and dig into the ontologies (and backing tools), check out our GitHub repository. The following folder structure is used:
- ontologies holds the definitions of the concepts and relationships that will be extracted from the narratives and online/structured data
- All of the files are written in Turtle (OWL2)
- ontol-docs contains documentation explaining the ontologies and their usage
- The graphs sub-directory contains PNGs of the main ontology concepts, serving as visual aids in understanding
- The file, dna-ontology-tree.html, holds a searchable tree view of the complete set of concepts and relationships
- notebooks holds the Jupyter notebooks used to scrape/parse web pages, and perform initial parsing and analysis experiments on the narratives
- dna contains the (evolving) Deep Narrative Analysis application executed through a simple GUI
- The GUI was developed using PySimpleGUI (documented at https://pysimplegui.readthedocs.io/en/latest/)
- No changes were made to the imported PySimpleGUI module
- tests holds pytest validation code for the dna application
- This code is NOT executed when pushing new code (as a github workflow) since a Stardog server would have to be deployed
- However, the code is run locally and the htmlcov sub-directory is included to show results
In the coming weeks, there will be many more posts on this blog regarding how to use and extend (and how we are using and extending) the ontology and tools. The goal is not to create a deep and complex hierarchy of concepts, but to create a logical infrastructure to describe a wide variety of events and situations, and then to extract these from narratives (text and speech).
Of course, the 'devil is in the details' of the specific domain of interest. OntoInsights is analyzing Holocaust narratives to illustrate our work. However, these ontologies and our approach can be generally applied.
Comments
Post a Comment