How to config spaCy NER model to stop tokenizing words and giving false positives?

Question

I am using spaCy's NER model "en_core_web_lg". This is the basic function that I have.

def find_pii_in_text(text):
    pii_dict = {}

    global nlp
    if not nlp:
        nlp = spacy.load("en_core_web_lg")

    clean_html = re.compile("<.*?>")
    text = re.sub(clean_html, "", text)

    text = re.sub(r"[^\x00-\x7F]+", " ", text)

    text_ent = nlp(text)

    pii_dict.update(process_flagged_data(text_ent, pii_dict))

    return pii_dict

def process_flagged_data(doc_list, pii_dict={}):
    for idx, entity in enumerate(doc_list.ents):

        # Ignore some labels

        pii_info = f"{entity.label_} : {entity.text}"
            if pii_loc not in pii_dict:
                pii_dict[pii_loc] = {}

            pii_dict[pii_loc].update(
                {
                    f"n{idx}": {
                        "text": entity.text,
                        "type": entity.label_,
                    }
                }
            )

    return pii_dict

Now, when I am passing sentences to this function, I am getting false positives in form of parts of words.

For example,

• the word 'movement' was flagged as 'movem' : PERSON.
• the word 'beginner' was flagged as 'begi' : PERSON.

Any idea as to why this is happening? Is there a config to tell spacy to not tokenize words?