Lost updates is a scenario which is studied when we look at transaction schedules. It has nothing to do with normalisation.
When data is deleted from unnormalised data, it could result in some copied of that data being left undeleted. However, this is probably not a "problem" as such.
Inserting data into an unnormalised system can result in the same data being entered in multiple tables. This gives rise to the possibility of data for the same object being entered "slightly differently" in more than one table. For instance, if a person's surname appeared in more than one table, and that person then changed his name, this change would have to be done in all tables where the name appeared. Ensuring this can be difficult, and so it is better to avoid such duplicated data by normalising the database design.
This is a way to implement transactions, and has nothing to do with normalisation.
Deadlock occurs when attributes are being modified in a system which uses locking to enforce serialisability. It has nothing to do with normalisation.
A lack of normalisation can lead to which one of the following problems