scripts/flagHallucinations.py

# Copyright 2024 FBK

# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at

#     http://www.apache.org/licenses/LICENSE-2.0

# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License

try:
    import pandas as pd
except ImportError:
    print("Please install the pandas package with 'pip install pandas' and try again.")
    exit(1)

import argparse

_VERSION = "1.01"

class ExplicitDefaultsHelpFormatter(argparse.ArgumentDefaultsHelpFormatter):
    def _get_help_string(self, action):
        if action.default is None or action.default is False:
            return action.help
        return super()._get_help_string(action)

# max size of pattern (seq of words) that,
# if repeated at least thresh1grams/threshNgrams times,
# risesflags the hallucination:
maxN = 5

def findHall(wrd):
    for N in range(maxN, 0, -1):
        count = 0
        for idx in range(0,len(wrd)-2*N+2): # the max idx value must be leave
                                            # room for at least one repetition
                                            # of the N_sized pattern 
            count += hallLen(idx,N,wrd)
            if ( N<3  and count+1>=parsed_args.thresh1grams ) or \
               ( N>=3 and count+1>=parsed_args.threshNgrams ):
                return [N, idx, count] 
            else:
                count = 0 # reset
                
    return [0,0,0]

def hallLen(startIdx,N,wrd):
    hallLen = 0
    startRep = startIdx + N # first index after the end of the current pattern
    while startRep < len(wrd)-N+1:
        if isHall(startIdx,startRep,N,wrd):
            hallLen+=1
            startRep+=N
        else:
            break
    return hallLen

def isHall(s1,s2,N,wrd):
    i = 0
    while i<N and wrd[s1+i] == wrd[s2+i]:
        i+=1
    return i == N
    
def main(args):
    """
    This script flags (by setting True the corresponding entry 
    of the hall_repeated_ngrams column) those sentences where a 
    pattern (n-gram, that is a sequence of n words) is repeated 
    at least a given number of times; for patterns of size 1 to 2, 
    the minimum number of times for flagging it is set by the 
    thresh1grams parameter (default value: 4), for those of size 
    3-5 by threshNgrams (2)
    """

    if (parsed_args.version): 
        print(f"Version {_VERSION} of anomalous string detector")
        exit(1)

    if not (tsv_files_specified):
        print("--tsv-InFile and --tsv-OutFile are both required")
        parser.print_usage()
        exit(1)

    if not (wrong_thresh_values):
        print("--thresh1grams and --threshNgrams must both be positive integers")
        parser.print_usage()
        exit(1)

    try:
        inDF = pd.read_csv(args.tsv_InFile, sep='\t', dtype=str, low_memory=False, na_filter=False, quoting=3)
    except IOError:
        print("Error in opening "+args.tsv_InFile+" file")

    try:
        txt = inDF[parsed_args.column]
    except KeyError:
        print("Error in reading column <"+parsed_args.column+"> in TSV file")
        exit(1)

    flag = []
    for line in txt:
        words = line.split()
        [size, idx, count] = findHall(words)
        if size>0:
            if args.quiet:
                flag.append("True")
            else:
                flag.append("True (pattern of length " + str(size) + \
                            " from index " + str(idx) + \
                            ", repeated at least " + str(count+1) + " times)")
        else:
            flag.append("False")

    inDF['hall_repeated_ngrams'] = flag
    inDF.to_csv(args.tsv_OutFile, sep="\t", index=False, quoting=3)


if __name__ == '__main__':
    parser = argparse.ArgumentParser(formatter_class=ExplicitDefaultsHelpFormatter)

    # I/O related arguments
    parser.add_argument(
        '--tsv-InFile', '-i', type=str,
        help="The input TSV file [Mandatory]")

    parser.add_argument(
        '--tsv-OutFile', '-o', type=str,
        help="The output TSV file [Mandatory. If equal to input TSV file, the new column is added to the original file]")

    # Processing arguments:
    parser.add_argument(
        '--column', '-c', default='source',
        help="Column name of the text to process [Optional]")
    
    parser.add_argument(
        '--thresh1grams', '-u', type=int, default=4,
        help="Threshold for 1-2_word hallucinations [Optional]")

    parser.add_argument(
        '--threshNgrams', '-n', type=int, default=2,
        help="Threshold for 3-5_word hallucinations [Optional]")

    # Reporting related arguments
    parser.add_argument(
        '--quiet', '-q', default=False, action='store_true',
        help='Print only True/False, no explanation for True\'s')

    # Get version information:
    parser.add_argument(
        '--version', '-v', action='store_true', default=False,
        help="Print version of the script and exit")

    
    parsed_args = parser.parse_args()
    tsv_files_specified = \
        getattr(parsed_args, 'tsv_InFile') is not None \
        and len(parsed_args.tsv_InFile) > 0 \
        and getattr(parsed_args, 'tsv_OutFile') is not None \
        and len(parsed_args.tsv_OutFile) > 0

    wrong_thresh_values = parsed_args.thresh1grams > 0 \
                          and parsed_args.threshNgrams > 0
    
    
    main(parsed_args)