Some differences in how lexemes are stored in v2.2 vs v2.3 models lead to this problem. We'll fix this in the next release, but if you'd like to patch your terms.teach recipe in the meanwhile, you can add the following around line 56 right after nlp = spacy.load(vectors) in recipes/terms.py:
for s in nlp.vocab.vectors:
nlp.vocab[s]