When constructing a conversational agent using IBM Watson Assistant, the starting activity is to gather representative questions. This helps to understand the types of interactions and queries the agent will need to handle, ensuring that the training data covers a broad range of typical user inputs. This foundational step is critical to effectively design and train the agent for accurate and relevant responses.

When you want to translate the language of text but are uncertain of the original language, you first need to identify the language. The IBM Watson Language Translator service provides an 'identify' endpoint specifically for this purpose. This endpoint analyzes the text and determines the most likely language the text is written in.

The IBM Watson image service that allows training based on custom images is Visual Recognition. This service provides capabilities for analyzing images, training custom models with specific images, and enhancing recognition performance based on the provided datasets.

Private documents can be uploaded to IBM Watson Discovery service after creating the collection. The collection is where the documents reside and can be managed. Therefore, it is essential to first have a collection to which the private documents can be uploaded.

The formula for recall in a classification system is True Positives / (True Positives + False Negatives). Recall measures the ability of a classifier to find all the positive instances and is thus calculated by dividing the number of true positive results by the sum of true positive and false negative results.