File size: 9,795 Bytes
6fa4bc9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 | {
"paper_id": "P81-1010",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T09:14:58.099847Z"
},
"title": "What Makes Evaluation Hard?",
"authors": [],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "",
"pdf_parse": {
"paper_id": "P81-1010",
"_pdf_hash": "",
"abstract": [],
"body_text": [
{
"text": "Ideally, an evaluation technique should describe an algorithm that an evaluator could use that would result in a score or a vector of scores that depict the level of performance of the natural language system under test. The scores should mirror the subjective evaluation of the system that a qualified judge would make. The evaluation technique should yield consistent scores for multiple tests of one system, and the scores for several systems should serve as a means for comparison among systems.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Unfortunately, there is no such evaluation technique for natural language understanding systems. In the following sections, I will attempt to highlight some of the difficulties",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The first problem is to determine who the \"qualified judge\" is whose judgements are to be modeled by the evaluation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "One view is that he be an expert in language understanding.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "As such, his primary interest would be in the linguistic and conceptual coverage of the system.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "He may attach the greatest weight to the coverage of constructions and concepts which he knows to be difficult to include in a computer program.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "Another view of the judge is that he is a user of the system. His primary interest is in whether the system can understand him well enough to satisfy his needs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "This Judge will put greatest weight on the system's ability to handle his most critical linguistic and conceptual requirements: those used most frequently and those which occur infrequently but must be satisfied. This judge will also want to compare the natural language system to other technologies. Furthermore, he may attach strong weight to systems which can be learned quickly, or whose use may be easily remembered, or which takes time to learn but provides the user with considerable power once it is learned. 3) How many hours of down time did plane 3 have in January, 1971 4) How many hours of down time did plane 3 have in February, 1971 will h~ip the success rate more than one query like, 5) How many hours of down time did plane 3 have in each month of 1971, However, ~ne query like 5 returns as much information as the other twelve. In testing PLANES (Tennant, 1981] , the users whose questions were understood with the highest rates of success actually had less success at solving the problems they were trying to solve.",
"cite_spans": [
{
"start": 649,
"end": 663,
"text": "February, 1971",
"ref_id": null
},
{
"start": 881,
"end": 896,
"text": "(Tennant, 1981]",
"ref_id": null
}
],
"ref_spans": [
{
"start": 5,
"end": 149,
"text": "Judge will put greatest weight on the system's ability to handle his most critical linguistic and conceptual requirements: those",
"ref_id": null
}
],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "They spent much of their time asking many easy, repetitive questions and so did not have time to attempt some of the problems.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "Other users who asked more compact questions had plenty of time to hammer away at the queries that the system had the greatest difficulty understanding.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "Another difficulty with success rate measurement is the characteristics of the problems given to users compared to the kind of problems anticipate~ by the system. I once asked a set of users to write some problems for other users to attempt to solve using PLANES.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "The problem authors were familiar with the general domain of discourse of pLANES, but did not have any experience using it.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "The problems they devised were ~easonable given the domain, but were largely beyond the scope of PLANES ~ conceptual coverage.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "Users had very low success rates when attempting to solve these problems. In contrast, problems that I had devised, fully aware of pLANES ~ areas of most complete Coverage (and devised to be easy for PLANES}, yielded much higher success rates. Small wonder.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "The point is that unless the match between the problems and a system's conceptual coverage can be characterlsed, success ~ates mean little.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PERSPECTIVE OF THE EVALUATION",
"sec_num": "2.0"
},
{
"text": "Testing a natural language system for its performance with with users is an engineering approach.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "4\u00b00 TAXONOMY OF CAPABILITIES",
"sec_num": null
},
{
"text": "Another approach is to compare the elements that are known to be involved in understanding language against the capabilities of the system. This has been called \"sharpshooting\" by some of the implementers of natural language systems. In this way, the capabilities of the system can be compared to the needs of the users. The relative importance of phenomena can be determined this way.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "4\u00b00 TAXONOMY OF CAPABILITIES",
"sec_num": null
},
{
"text": "Furthermore, users\" language can be studied without them adapting to the system's limitations. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "4\u00b00 TAXONOMY OF CAPABILITIES",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {},
"ref_entries": {}
}
} |