Difficulties of computers conversing using human languages
Daily, people take advantage of the work done by computer linguists, on multilingual hotlines or online dictionaries, just to name a few. The real challenge for the people developing these technologies is closing the gap between the formal logic of computers and the complex natural languages we speak. The field of computer linguistics grew with rudimentary machine translations back in the 1950s and 60s.Since the 1980s, lots of progress has been made in the field and many language technologies are still evolving at a rapid pace. A challenge that has remained in developing these technologies is the use of formal mathematical logic to help computers understand the phenomena of dynamic, natural and situation-dependent language.
Various factors explain the difficulties computer linguists have encountered in ‘teaching’ machines to converse. Firstly, the problem of speech synthesis and slow reaction times (RT) arises. At word level, speech synthesis is quite effective as artificial voices can match human ones. At phrase and sentence level, issues regarding intonation emerge resulting in speech that sounds quite unnatural. This is a result of monotone the voices used as well as varying rhythm used in different situations. As a result, communication is slowed for the sake of accuracy.
Human language is unconscious in expression and communication depends on situations as concerned. A word may have varying applications relying on the scenario. For
example, a human may mention the word ‘swell’ in happiness. A machine however may register the meaning as a growth of sorts! These discrepancies lead to ambiguity after processing of speech by the aforementioned machines. Wrong lexical decisions result in unsuitability for application (of these machines) in varying fields other than those using specific syntax.
Machines have been limited to select responses to human queries. This inflexibility is a limit to productive. In automated call-centers, customer issues are not addressed directly but instead routed through various steps such as ‘press 1…’e.t.c and eventually to a call-care representative. Productivity is thus lost on operational costs and time on the customer’s side. These limitations exist due to the sheer number of algorithms that have to be created to cope with varying customer queries and needs.
Syntactic Ambiguities are a major problem in teaching computers to converse using human languages. Wrong syntactic structure of a word set by a human may cause a successful algorithm set in a machine to fail entirely. Ambiguous results or none then ensue. It is therefore imperative that a preset syntax is followed to the letter in use of these systems. Educating users on proper syntax of queries is an expensive matter.
Variability in human expression is a major problem for computer linguists. It is further accentuated by use of various superlatives. Processing of such language takes very different paths. Output from such machines may appear ambiguous or even offensive to humans depending on the path taken. This discourages companies from investing in such technologies or limiting the scope of their use.
Machines usually have a limited knowledge base. Human queries are often met with no replies whatsoever due to no processing taking place.
Fromkin, Victoria, and Robert Rodman. An Introduction to Language. New York: Holt, Rinehart and Winston, 1978. Print.